Home

Master thesis describing version 0.98 - GRC

Contents

1. Source Folder home wannesjtest2 Search Resuts Folder homewannesitest2 search New Simulation Load Scenario Import NS 2 Scenario import cityMob Scenario Edt Simulation Repetitions Fk _E Delete Simulation Load list Save list Configuration File Content LnkQualityAlgorithm etx fom Interface lar U IpdBroadcast 255 255 255 255 Hellointerval 2 0 Hellovelidityrime 40 0 Telnterval 5 0 TeValidityTime 100 0 Midinterval 18 0 MidVelidityTime 224 0 Hnainterval 18 0 Hnavalidtytime 108 0 Path protocol conf fetc olsrd conf nnodesfilf checkPoint NodeCheckPoint checkPoint new NodeCheckPoint float p x float py new Float 0 myelugin Generate Plug in Name Name age shomejwannes router Install router Delete TE Dele S S Folders f Pom Xam 0m 800 mm OT Ido Simulation Corfiguratien Help Y Simulation S RE nnn inetbeans 3 8 S meen pee Access Point New iN ezn Control IP Adress Control IP Adr TEE ran lt Load CES laea 192 168 1 202 192 168 1201 Ping Laan J
2. Source ddress Transfer Size Throughput Delayims lt no fso ene Irouter2 UDP 512 ja 200 bo 00 o false a no 20 routera routers Ter 1009000 5 o aj 60 110 router routers TCP 1000000 5 a T o I L K routerz routera UDP 1000 512 a 400 Do 00 bo false a Duplicate Row Delete row Order traffic Delete all Status Upp packets n veed 0 w waiting Average Throguhput o brs Save intex file close STADIVA RAN 250 m 1006 ml Sources _ S Start is Stop 5 Traffic Transfer Size Size ofpacket _ Pkts Secand Flows oe so lupe l 1400 10 H EM K lube 1400 10 1 Nodes El E Granularity Runs S R St Order traffic Duplicate row Delete row Delete al Simulation a ER It Size Traffic Flow Protocol Used Save in Falder Export format rar un Balanced Z None aopy Select castadiva Random O Ns 2 L E E 7 optimum C OLSR New Folder File Rename File A00 en AP
3. 18 Objectives 20 Results 21 6 1 How to setup the test bed 21 61 15 Setup the Core asz sz a ao ia Bleek RUN a pen Rennes ee ces 21 6 12 Setup the Nodes Legs JE ge hae ape he pat tele b 22 6 1 3 Compile for the routers 24 6 2 Validation of new functionalities 25 6 2 1 The execution planner 25 6 2 2 Routine Plugins sentsa oe w s mas ban sh bea ee be a 28 6 2 3 Mobility Plugins 35 6 2 4 CityMob Import 39 6 2 5 Minor issues in Castadiva 44 6 3 Important mobility issues in Castadiva 44 6 3 1 Bad throughput calculation 44 6 3 2 Bad ARP emulation with TCP 46 6 4 Maximal UDP bit rate issue 47 6 5 Writing a research Article 49 7 Useful software 50 Tole NEDES or nue T tn PR TR ane ot be eines R R LONG ee A m0 50 7 2 Network Simulator 2 NS 2 50 7 21 Introduction e amp 44 2 5 eux wR a Bone duo da die t 50 7 2 2 Learning about ns 2 51 7 2 3 Simulating 802 11g with ns 2 aoaaa aa a 51 7 2 4 Setting ranges with ns 2 51 7 2 5 Analyzing ns 2 results 52 1 2 6 Casta
4. gt set val nn_session val nn_session puts gt set val interSession val interSession puts gt set val interPageOption val interPageOption puts gt set val pagesize val pagesize puts gt set val objsize val objsize puts n n n F SEE Main Program ZEEZEZEEEZEEEZEEEEZ EE SS EE SS EE E EE EE E EE EE E EE Initialize Global Variables 72 LL set mindelay_ 50us LL set delay 25us LL set bandwidth __ 0 not used LL set off prune 0 not used LL set off CtrMcast_ 0 not used Agent Null set sport_ 0 Agent Null set dport_ 0 Agent CBR set sport_ 0 Agent CBR set dport_ 0 Agent TCPSink set sport_ 0 Agent TCPSink set dport_ 0 Agent TCP set sport_ 0 Agent TCP set dport_ 0 Agent TCP set packetSize 1460 give preference to routing protocols Queue DropTail PriQueue set Prefer Routing Protocols 1 unity gain omni directional antennas set up the antennas to be centered in the node and 1 5 meters above it Antenna OmniAntenna set X_ 0 Antenna OmniAntenna set Y_ 0 Antenna OmniAntenna set Z 1 5 Antenna OmniAntenna set Gt_ 1 0 Antenna OmniAntenna set Gr_ 1 0 Initialize the SharedMedia interface with parameters to make it work like the 914MHz Lucent WaveLAN DSSS radio interface Phy WirelessPhy set CPThresh_ 10 0 Phy WirelessPhy set CSThresh_ 1 559e 11 Phy WirelessPhy set RXThresh_ 3 652e 10 Phy WirelessPhy set Rb_ 11 1e6 Phy WirelessPhy set Pt_ 0 2
5. node 1 set Z 0 0 Node 2 210 11858578806059 560 0 gt 1240 0 560 0 59 38569098715311 Km h node 2 set X 210 11858578806059 node_ 2 set Y_ 560 0 node 2 set Z 0 0 Node 500 0 145 20410617171447 gt 500 0 380 0 67 02568486533434 Km h node 3 set X_ 500 0 node 3 set Y 145 20410617171447 node 3 set Z 0 0 Node 4 800 0 693 2904817726454 gt 800 0 340 0 59 76214259985875 Km h node 4 set X 800 0 node 4 set Y 693 2904817726454 node 4 set Z 0 0 END OF INITIAL POSITION CONFIGURATION Movements NODE MOVEMENTS Node 0 had an ACCIDENT ns_ at 0 0 node 91 setdest 1140 0 1390 6241869639089 0 0 END ACCIDENT NODE 1 MOVEMENTS Node 1 TARGET REACHED 540 0 1200 0 75 06782928526441 Km h ns_ at 52 18759071966314 node 1 setdest 540 0 1200 0 75 06782928526441 Node 1 NEW TARGET 40 0 1200 0 Node 1 TARGET REACHED 40 0 1200 0 52 77465990538741 Km h ns at 86 30727806161097 node 1 setdest 40 0 1200 0 52 77465990538741 Node 1 RED SEMAPHORE UNTIL 104 4088489660594 Node 1 GREEN SEMAPHORE Node 1 NEW TARGET 40 0 520 0 ns_ at 104 45 node 1 setdest 40 0 1200 0 0 0 Node 1 SPEED CHANGE 0 0 Km h gt 63 40445188023556 Km h ns at 120 0 node 1 setdest 40 0 926 1279925728561 63 40445188023556 NODE 2 MOVEMENTS Node 2 TARGET REACHED 1240 0 560 0 59 38569098715311 Km h ns_ at
6. This was suggested by Carlos Those packets where then re routed over a wireless link We got only 2Mbps throughput We did the same with an EEEPC as a server connected to the second router Again we got only 2Mbps with UDP Da g EEEPC Automatic Router Client packet forward Server to Server Figure 6 19 Test architecture for UDP routing on wireless routers 5 We found out that the processor of client routers do rapidly saturate with UDP transfers Optimizing the client source code and optimizing the compilation we got a slightly better throughput with UDP But still much lower than TCP s throughput and processor saturation As a conclusion it seems that LinkSys WRT54GL routers have problems handling wireless UDP packets At least with OpenWRT 6 5 Writing a research Article It was an unsuspected task On April the 20 I was asked to participate in a research article in collaboration with Alvaro Torres The article comments the new functionali ties in Castadiva and it is supposed to be presented during the CEDI 2010 congress in September A copy of the final article can be found in appendix B As it was a relatively urgent task I partly suspended my previous work to dedicate about one week to the writing of the article In fact writing was probably just a minor task even though we had to write in English What really gathered my efforts was to get a clear idea of Castadiva my work and what I was going to wr
7. Video AC_ VI Best effort AC_BE and Background AC_ BK For applications to take advantage of the IEEE 802 11e technology datagrams should have their IP Type of Service TOS header field set according to the desired user priority When delivered to an IEEE 802 11e enabled wireless card driver those datagrams will be handled according to the priority defined In Castadiva we have implemented the packet priority support by adding TOS infor mation to UDP packets in our generator Fig ure 5 shows how the user can select one of the four categories for each UDP flow If the subjacent network is not IEEE 802 11e enabled setting the TOS won t make any difference however if the network is IEEE 802 11e enabled traffic differentiation should be effective mtrol cts Sec Max Pa Throughput Pkts Re fMean Acce IDACME Delay Redir 400 0 0 0 0 0 0 AC VO false false 100 500000 0 0 0 0 0 0 AC BE false 400 0 0 0 0 0 0 Ac false false 0 0 A M 400 0 0 0 0 0 ais false AC_BE ACM AC VO UDP Packets Reveived 0 l Save in text file Average Throguhput 0 kb s Close Figure 5 IEEE 802 11e category selection and De lay measurement 65 66 CASTADIVA TRAFFIC JE Traffic Control t Start 5 Stop 5 Source Address Traffic 10 110 JAP1 JAP2 JUDP 10 110 AP3 AP2 TCP Transfer Size B _ Size of packet
8. amp Jorge Hortelano Master and PhD students in the GRC for their pa tience an their answers to many of my questions about Castadiva 99 Index A O Alvaro 9 Objectives 12 20 OLSR 29 C OpenWRT 17 22 Canvas 18 OPTIMUM 28 Castadiva 15 CBR 46 P CEDI 2010 49 Pietro Manzoni 10 CityMob 39 Core 16 R Cross compiling 25 RO PRES 2g S E SSH 17 ERASMUS 8 SSH Public Key 24 STRI 3 Gnuplot 52 T GRC 9 TFTP 23 I U IDE 50 UPV 9 J V jar 21 VANET 39 JAVA 17 javac 21 Jorge 9 L LATEX 53 LinkSys WRT54GL 17 LyX 53 M MANET 15 Marta Caballero 10 Mobility Plugin 35 N Netbeans 50 Network Simulator 2 ns 2 50 NFS 17 21 Node 16 56 Bibliography 1 Jorge Hortelano Juan Carlos Cano Carlos T Calafate and Pietro Manzoni Testing applications in manet environments through emulation EURASIP Journal on Wire less Communications and Networking vol 2009 Article ID 406979 20 pages 2009 doi 10 1155 2009 406979 2 Wannes Vossen Alvaro Torres Jorge Hortelano Carlos T Calafate Juan Carlos Cano and Pietro Manzoni Extending an emulation platform for automatized and distributed evaluation of qos in manets April 2010 57 Part III Appendices A Castadiva s GUI splash view 09 ADVA ETRAF Traffic Control
9. gt gt javac c gt i E gt jar p G MobilityPlugin l i Wrapper java plugin class plugin jar MobilityPlugin Wrapper txt Figure 6 8 Creation of a jar ServiceLoader Mobility Plugin Important note The automated compilation uses certain files in the source src folder of the project Make sure that the following constants are correct in Castadiva Model java e PLUGIN_ WORKFOLDER pluginTemporaryFiles is the name of the temporary folder that is created for compilation purpose e MOBILITY_PLUGIN_JAR_FOLDER src castadiva MobilityPlugins is the path to the Plugin folder where jar files are copied after compilation and where Plugins are loaded on startup e PLUGIN_INCLUDE_FOLDER src lib must point to the folder containing the IPluginCastadiva java abstract class 6 2 3 3 Usage The Mobility Plugin System is quite simple to use when it is correctly understood Fig ure 6 9 shows the unique user interface that can be used to generate a custom mobility plugin In the upper part of the figure we can see the header of a JAVA function The variables shown in that header are the one that the user can use in order to design his algorithm The only requirement for the final algorithm is that it entirely fills the NodeCheckPoint nodes matrix In order to create a NodeCheckPoint instance the NodeCheckPoint class can be used as follows NodeCheckPoint checkPoint new NodeCheckPoint float XPosition float YPos
10. proc sys net ipv4 ip_forward iptables t nat A POSTROUTING o ethO j MASQUERADE e opkg is the packet manager in the OpenWRT systems e etc rc d S90custom user startup sh is OpenWRT s specific file for user commands at startup e The pub file is a public key for SSH It is recommended to configure Open SSH public key authentication among your Core and your routers That will prevent you entering a password each time you connect to a router To generate a public key for root on the core computer do the following as root ssh keygen t rsa Set no paraphrase e The files in etc config are used to properly configure the router Those files are read at router s start up and overwrite any common Linux configuration I would recommend to copy and configure those files from the first updated router as they can change from one to another version 6 1 3 Compile for the routers Now that the routers are correctly configured there is one last step we have to fulfill to obtain a Castadiva ready system In order to simulate traffic among nodes Cas tadiva uses binary files and runs them directly on the nodes Those files are simple TCP and UDP clients and servers written in C We thus need to compile those files for our 3 scp allows to transfer files over ssh r stands for recursive it allows to copy entire folders The source files should be provided with the program they should be located in a bin directory 24 Ope
11. puts gt set val mac Mac 802 11 MAC type puts gt set val ifq Queue DropTail PriQueue interface queue type Queue DropTail PriQueue for AODV CMUPriQueue for DSR puts gt set val ifq CMUPriQueue for DSR puts gt set val ll LL link layer type puts gt set val ant Antenna OmniAntenna antenna model puts gt set val x val x X dimension of the topography puts gt set val y val y Y dimension of the topography puts gt set val ifqlen 50 max packet in ifq puts gt set val seed 0 puts gt set val rp val rp one of AODV AODVUU DSR TORA OLSR v7 puts gt set val nn val nn how many nodes are simulated puts gt set val nn traffic sources val nn traffic sources nodes que navegaran por web puts gt set val cp val cp patron de conexiones en web no utilizar puts gt set val sc val sc escenario de trafico CAMBIAR puts gt set val stopns val stopns simulation time puts gt set val trfile val trfile puts gt set val namfile val namfile puts gt set val httpfile val httpfile puts gt set val random mobility val random mobility puts gt set val random number generator val random number generator puts gt set val bw val bw puts gt set val completion val completion puts gt set val nn_pages val nn_pages puts
12. tems Local and Metropolitan Area Net 1 Network Time Protocol works Specific Requirements Part 11 2 3 4 5 6 http datatracker ietf org wg ntp charter Acessed May 3 2010 J C Cano J M Cano C Calafate E Gonzalez and P Manzoni Evaluation of the trade off between power consump tion and performance in bluetooth based systems Sensor Technologies and Ap plications International Conference on 0 313 318 2007 Charles E Perkins Elizabeth M Belding Royer and Samir R Das Ad hoc on demand distance vector AODV routing Request for Com ments 3561 MANET Working Group http www ietf org rfc rfc3561 txt July 2003 Work in progress D Raychaudhuri I Seskar M Ott S Ganu and K Ramachandran Overview of the orbit radio grid testbed for eval uation of next generation wireless net work protocols Wireless Communica tions and Networking Conference 2005 IEEE 3 1664 1669 2005 Glenn Judd and Peter Steenkiste De sign and implementation of an rf front end for physical layer wireless network emula tion In IEEE 2007 IEEE 65th Vehicular Technology Conference VTC2007 April 2007 IEEE 802 11 WG International Standard for Information Technology Telecom 8 10 Wireless Medium Access Control MAC and Physical Layer PHY Specifica tions ISO IEC 8802 11 1999 E IEEE Std 802 11 1999 IEEE 802 11 WG 802 11e IEEE Standard for Information technology
13. that command line is hard coded what if it changes with different nodes or different oper ating system versions Therefore a plugin system was developed and implemented in Castadiva The idea is to allow the user to define the needed informa tion to manage routing on the different nodes Moreover it offers the advantage of possibly designing a routing protocol plugin for any routing protocol supported by the nodes B Configure Protocols gt New Protocol Configuration Name oLsr Bin jusr sbinyolsrd A Flags L Configuration File Content Fs al olsr org OLSR daemon config file M Lines starting with a are discarded This file was shipped with olsrd 0 X X This file is an example of a typical configuration for a mostly static network regarding mobility using the LQ extention Debug level 0 9 Path protocol conf Jetcjolsrd conf C Create plugin Cancel Figure 6 Routing plugin configuration window in Castadiva Figure 6 shows how routing plugins can be configured in the current version of Castadiva The command in text field A is sent to each access point when the simulation starts fol lowed by eventual flags All text entered in text area B is saved as a file that is copied on the nodes at the path location stated in text field C All existing files are overwritten and the configuration file is removed after the sim ulation Finally in order to find out h
14. we need a computer with the following characteristics e One Ethernet network card No Wireless Card needed e Support of Network File System NFS as a Server e Support of JAVA to run the Core program e Support of the javac JAVA Compiler and jar Creation of a jar archive commands to use Castadiva s plug in system In our example we use a PC with a Debian distribution Here is how to install the NFS Server and how to export a directory 1 Install the NFS Server 21 apt get install portmap nfs common nfs kernel server dpkg l portmap nfs common nfs kernel server dpkg reconfigure portmap w N 2 Create and export the CASTADIVA directory mkdir CASTADIVA 2 echo CASTADIVA 192 168 1 0 255 255 255 0 rw sync no_subtree_check no_root_squash gt gt etc exports 3 exportfs ra 3 Configure the hosts deny and hosts allow files cat etc hosts deny lockd ALL mountd ALL rquotad ALL statd ALL oR WN cat etc hosts allow lockd 192 168 1 1 192 168 1 201 rquotad 192 168 1 1 192 168 1 201 mountd 192 168 1 1 192 168 1 201 statd 192 168 1 1 192 168 1 201 oR WN 6 1 2 Setup the Nodes It is mandatory that the nodes run a Linux based operating system moreover they should be able to e Support C compilation A cross compiler might be a solution e Support Network File System NFS as a client e Have two network interfaces One supporting Ethernet and the being wirel
15. 11 Mac 802 11 Mac 802 11 Mac 802 11 Mac 802 11 Mac 802 11 set set set set set set set set set set set set CWMin_ 15 CWMax_ 1023 SlotTime_ 0 000009 SIFS_ 0 000016 PreambleLength_ 96 PLCPHeaderLength_ 40 PLCPDataRate_ 6 0e6 RTSThreshold 3000 ShortRetryLimit_ LongRetryLimit_ basicRate_ 6e6 dataRate 54e6 AN The RX treshold determines the range Processed with threshold cc Phy WirelessPhy set RXThresh_ 9 90556e 10 32dBm Phy WirelessPhy set Pt_ 1 58489 Create and configure topography used for set topo new Topography 1000x1000m terrain topo load flatgrid 1500 1500 Configure the future nodes ns node config adhocRouting val rp LlType val 1l macType val mac ifqType val ifq ifqLen val ifqlen antType val ant propType val prop phyType val netif channel new val chan topoInstance topo agentTrace OFF routerTrace OFF macTrace ON ifqTrace OFF for 802 11a and 802 11g 20us 10us 144 bit 48 bits 1Mbps bytes retransmittions retransmissions mobile scenarios 79 F Configuration for 802 11g in ns 2 with distance losses GO J O U1 E UKM KA Default NS2 header for Wannes s work on Castadiva s validation Simulates a 802 11g wireless network Parameters from Yojan Version 1 2010 04 16 KA KK K K K K OK OK K K K K K K K OK K OK 2K OK K OK K K OK OK OK OK K K K K K K
16. 46 22910709632431 node 4 setdest 1440 0 340 0 92 42208279251665 Node 4 RED SEMAPHORE UNTIL 55 864743625517406 Node 4 GREEN SEMAPHORE Node 4 NEW TARGET 1440 0 120 0 ns_ at 55 9 node 4 setdest 1440 0 340 0 0 0 Node 4 SPEED CHANGE 0 0 Km h gt 79 91145948724842 Km h Node 4 TARGET REACHED 1440 0 120 0 79 91145948724842 Km h ns_ at 65 8109690284956 node 4 setdest 1440 0 120 0 79 91145948724842 Node 4 NEW TARGET 20 0 120 0 ns_ at 120 0 node 4 setdest 44 37136896264302 120 0 92 78417491661162 EXITING PROGRAM 76 E Configuration for 802 11g in ns 2 without distance losses O J nU E UKM KA ESS set val chan Channel WirelessChannel channel type set val prop Propagation TwoRayGround radio propagation model set val ant Antenna OmniAntenna Antenna type set val ll LL Link layer type set val ifq Queue DropTail PriQueue Interface queue type set val ifqlen 50 max packet in ifq set val netif Phy WirelessPhy network interface type set val mac Mac 802 11 MAC type set val rp DumbAgent ad hoc routing protocol set v l in 5 number of mobilenodes ZEEZEZEEZEZEEZEE ee Simulation SS Ee eee Create simulator set ns_ new Simulator Set up trace file ns use newtrace set tracefd open simple tr w ns_ trace all tracefd Setup Xgraph files set TU open out0 tr w set f1 open outl tr w set f2 open out2 tr
17. Configure your computer s network card to 192 168 1 x Where x 1 This is important the router will be automatically configured to 192 168 1 1 after a successful firmware update and you would have an IP conflict Unplug the power cord of the router and make sure the router is connected to the computers network Using a switch for example Run your tftp program and run the following command tftp gt connect 192 168 1 1 t tp gt mode octet t tp gt trace t tp gt timeout 1 gt tftp put openwrt wrt54g squashfs bin Immediately plug the router s power cord The transfer should start soon You might have to repeat those steps a few times before you succeed You may want to vary the time between the last tftp command and the router s start up If you do not succeed with the TF TP method It might be because of a previous wrong installation as stated in step 1 If this happens I recommend to use the mtd command line method descri We bed in the official OpenWRT wiki now need to establish a first connection using telnet in order to set the root password on the router Then we will need to configure the routers for Castadiva As I had 5 routers to manage and as I have had to reinstall routers several times to obtain a clean configuration I designed a script shell that automates the installation of the router That script shell is a good example to understand how to configure a router bin sh CASTA
18. OK OK K 2K 2K 2K K 2K 2K K K K OK K OK OK OK OK Define general simulation options DSR TORA OLSR v7 PR Tk E EE Ke eR HK h k HAN E KEE hh E E OK eR KA sensing threshold in dB above noise power set sensingTreshdB 5 set system type exec uname s Libraries have different names in some operating systems if string match CYGWIN system type 1 load usr bin ns allinone 2 34 dei80211mr 1 1 4 src libs cygdei80211mr 0 d11l else load usr bin ns allinone 2 34 dei80211mr 1 1 4 src libs libdei80211mr so set val chan Channel WirelessChannel PowerAware channel type set val prop Propagation FreeSpace PowerAware radio propagation model set val ant Antenna OmniAntenna Antenna type set val ll LL Link layer type set val ifq Queue DropTail PriQueue Interface queue type set val ifqlen 10 max packet in ifq set val netif Phy WirelessPhy PowerAware network interface type set val mac Mac 802_11 Multirate MAC type set val rp DumbAgent one of OLSR AODV AODVUU set val nn 5 number of mobilenodes set PHYDataRate Mode54Mb Z RAR HOME h hh A eK AE A eK He A E a A oe MEANS 2 Ok RES 2 a ok A ae ak HN Define advanced options KK KK OK K K K OK hh 2K K OK K OK K OK OK OK K OK K OK K OK K K K K K K K K OK OK K 2K 2K 2K K 2K K KK OK K K OK OK OK OK LL set mindelay_ lus LL set delay_ lus LL set bandwidth_ 0 not used Node MobileNode instproc getlfq param0
19. Olmos Filippo Squillace Figure 2 1 Organization chart of the GRC research group 2 3 An internship opportunity On exchange one should stay as close as possible to the material he would have studied in his home university My classmates in France had to perform a four month internship in some company On the occasion of an exchange it is quite common for a student to work in a university lab When I created the exchange agreement I asked my Spanish exchange coordinator Marta Caballero for such a placement and she put me in touch with Pietro Manzoni coordinator of the Computer Networks Group and also dealing with exchange in the computer science department A few emails later we had an overview of what I could possibly work on 2 4 Technical aspects On my arrival I was assigned to a desktop in the network research lab The material and the environment were modern and quite comfortable to work with We were more or less 6 Master and PhD students working int the laboratory and the atmosphere was pleasing One could easily ask another for advice and everybody was always pleased to help 10 Figure 2 2 Working material with the stacked routers 11 3 Objectives amp time management 3 1 Objectives Figure 3 1 shows how I spread my different tasks over the whole internship Tasks will be explained with more detail in Part II For now I will focus on time management and objectives On my first day I met Pietro Manzo
20. Telecommunications and information ex change between systems Local and metropolitan area networks Specific requirements Part 11 Wireless LAN Medium Access Control MAC and Physical Layer PHY specifications Amendment 8 Medium Access Control MAC Quality of Service Enhancements 2005 Jorge Hortelano Juan Carlos Cano Carlos T Calafate and Pietro Manzoni Testing applications in manet environ ments through emulation EURASIP Journal on Wireless Communica tions and Networking vol 2009 Article ID 406979 20 pages 2009 doi 10 1155 2009 406979 T Clausen and P Jacquet timized link state routing protocol OLSR Request for Comments 3626 MANET Working Group http www ietf org rfc rfc3626 txt October 2003 Work in progress Y Zhang and W Li An inte grated environment for testing mo bile ad hoc networks Available at http www wins hrl com projects adhoc Op 69 C Johann M Marquez Barja s implementation for 802 11g in ns 2 70 OONDUBWNEH HTTP en VANETs Simulation script for NS2 Johann Marquez johann marquez barja com puts gt Top Down Scripting n ZEEZEZEEEZEEEEEE SSS SS SS SS EE EE SSS SS SSS EE SS SS SSS SS E E SS SS SS SS E SS SS SSS SS Define options EEE EE set val chan Channel WirelessChannel channel type set val prop Propagation TwoRayGround radio propagation model set val netif Phy WirelessPhy network interface type set val m
21. also effective for professional edition Later on I discovered LyX a Graphical User Interface for ATEX which makes ATEX writ ing much more productive even tough it probably also introduces some limits I used ATEX with LyX for this report and got fully satisfied LyX handles many graph ical file formats which allows to insert vectorized drawings directly into a pdf document It offers to generate fully operational pdf documents with hyperlink and index management Finally YX also handles appendices with great effectiveness 53 8 Conclusion I did not achieve all of the tasks I was given for this internship At least it did not succeed in positively validating all of Castadiva s new options When we defined my work we did not know exactly how hard it would be It finally came up to be much more difficult than expected the functionalities that I had to validate were in a beta stage On one hand the code was chaotic sometimes Spanish written not commented and contained errors omissions On the other hand the previous designs were incomplete as they did not take in account all parts of the program For sure It is not an easy task to enable mobility in Castadiva which was initially designed without mobility The complexity makes it quite difficult and long to assimilate a new functionality It is also hard to find the origin of an error which can be as vague as wrong results Nevertheless regarding the task I did not entirely validat
22. by EDCA enhanced distributed channel access With this new function fully compatible with DCF QoS is achieved through the introduction of different access categories ACs and their associated backoff entities Contrarily to the legacy IEEE 802 11 sta tions where all packets have the same prior ity and are assigned to a single backoff entity IEEE 802 11e stations have four backoff enti ties one per AC so that packets are sorted ac cording to their priority The different access categories available in IEEE 802 11e stations CASTADIVA SIMULATION Time pe N Simulation Waiting s Z x arouter3 ri A om 200 m 00 m 600 m 0m 008 m 1200 m 1400 Simulation Wh pet S SI i i Access Point Bounds froute2 ae sz 7 X m C EZ m Position 15000 m x o m LY Ys m wa 265 Le ke ls WiFi Connection e _routerd Range 400 0 m Node Mobility K O RTS CTS Show maxspeed 20 00 ms Pause 00 s N PA Mobility Model I View M Grid RANDOM way POINT lv _ 6 Labels 1060 m O Movement Routing protocols L folsr x n FE New Simulation Replay Simulation 1400 m Stop Simulation Show Save Scenario Random Scenario Drawing area 1500 l s Cm CT Reset all AP lad J 500 1500 2500 3500 4500 5500 6500 7500 8500 9500 m Help Traffic Declare traffic Status Ready goss Figure 3 Simulation window are Voice AC_VO
23. integrates advanced functionalities the ones I used most are e Code completion based on the imported classes e Search for all the usages of a function or variable in the project e Search for the declaration of a function or variable in the project e Support of Subversion a version management system to enable effective collaborative work e Live error correction e WYSIWYG graphical user interface editor e Automated variable and function renaming For example when a I had to investigate a new functionality I systematically searched the corresponding code using the Find usages or Go to source options When the code was written in Spanish I could simply rename variables and functions in the whole project Having a list of available functions for a variable is very handful to handle very large classes The subversion system that was set up for Castadiva was also very well integrated in Netbeans It was easy to compare versions of the programs keep track of all the changes and release new versions of the software 7 2 Network Simulator 2 NS 2 7 2 1 Introduction The Network Simulator ns 2 is developed by various researchers and institutions It is frequently used in network research and was appropriate in our situation to compare Cas tadiva s results to other results In this section we wont describe all of ns 2 s possibilities but review use full tips for ns 2 with Castadiva MANET s and routing in general 1Se
24. reviewed Temporary folders are now deleted after each plugin creation 3 The initial plugin system did not remove any temporary file As a result any plugin contained all the files generated for all the previously generated Plugins Currently a single folder PLUGIN_WORKFOLDER is created and removed for each plugin creation 4 The initial plugin system was unable to handle any configuration file of more than one line It was also unable to handle odd symbols As each plugin contained all of his ancestors see 3 I could find out that the typical asdfasdfasdfasdf chain was recurrently used to test the Configuration File Content text area Any carriage return crashed Castadiva This was solved using a separate protocol conf file to store the protocol configuration data I also included the protocol conf file into the jar archive in order to have a single file for a plugin Moreover it allows to easily modify the configuration file 5 The code was partly Spanish written and there was practically no comment We can suppose that it was still under development I translated the code and wrote comments for the whole routing plugin functionality 6 The Flags text filed was not used in the program It is now usable 7 The configuration field was removed by the program after the simulation If the simulation failed the configuration file was not reset As for any other simulation element I planned the configuration file removal usin
25. 1 Receiving threshold RXThresh_ is 1 95665e 10 I made a few experiments with nodes separated with 300 299 9 or 301 to confirm that result 7 2 5 Analyzing ns 2 results I first used a method from Evan Jones s blog to collect UDP simulation results It consists in recursively running a function with a fixed interval of time and to write temporary results to a text file This was quite efficient for UDP and other data like distances between nodes or node s position but I could not translate the technique to TCP Carlos Miguel Calafate is the one who taught me an interesting method based on ns 2 s default output files and simple Script Shell scripts For example the following command cat simple tr grep r grep Md 4 grep Msi2 grep cbr we 1 Will return the number of UDP CBR packets that were transmitted from node 2 to node 4 There are of course many other possibilities and this method allows a deeper analysis and better understanding 7 2 6 Castadiva and ns 2 Castadiva offers two ns 2 related functions e import NS 2 was probably meant to offer an Export to Castadiva functionality in other network related software As an input it takes a very precisely formatted file Any space should be respected and a special header is also requested nodes 5 pause 0 0 max speed 0 0 max z 1500 0 max y 1500 0 e export NS 2 function is supposed to allow traffic ans scenario export for ns 2 For now TCP
26. 10 11 12 eneral review of Castadiva tudy of CityMob for ompatibility with Castadiva investigation and adjustment of he Routing plugin system Castadiva validation Tests with investigation and adjustment of e Import from CityMob option articipation to a Research article Investigation of a UDP hroughput issue Investigation and adjustment of e Mobility plugins system ng Figure 3 1 Task distribution over time 13 Part II Achieved work 4 What is Castadiva It is mandatory for further reading to understand what Castadiva is In Section 4 2 I will make a short description of it Please note that a more extended description of Castadiva is also available it can be read in the Castadiva Journal 1 4 1 MANETs Mobile ad hoc networks MANETs are packet radio networks composed by independent and heterogeneous stations that cooperate in routing and packet forwarding tasks All the nodes which are part of a MANET can act as routers allowing communication among several out of range end points As a MANET s nodes can also be moving around the the network topology can suffer constant changes In such situation dynamic routing protocols are very interesting to automatically reconfigure routes Testing and evaluating MANET management protocols is important to guarantee their effectiveness in real situations Researchers have three options to do so to use simulation t
27. 168 1 201 1021682201 l C sa 1500 2500 2500 4500 5500 6500 7500 esoo esoo m EET isease 132__ lame x Dent 6080 Help H Gateway SSID Traffic T Protocols FT rrr 11T Status Ready close i 7 Mobility Designer gt EE Pe SS ses va l maT lan Jen De ra Close H gi A NFS folder in the computer E ICASTADVA eases AP scripts destination folder gi L Le NFS Folder in the AP a frsstadvonts TGM AGE Mobility Model Designer SSH connection ed SSH user SSH pwd New Protecol Configuration e 9 HEC CE ea TI void ObtairNodePositionsForEntiresimulation NodecheckPoint nodes APs accessPoints Float minSpeed froot D Current P Name eler Float maxSpeed Float simulationPause int totalTime float X float Y uii 19216811 Source folder JPesuts folder Tune status Es e RATES r lint 1s lt totalime HT home wannesitest L ihome wannesjtest L 10 Ready Flags ICASTADIVA d pue the ELA ata nai che homeannes 10 Read AP p accessPoints Gent B Extending an emulation platform for automatized and distributed evaluation of QoS in MANETs 2 62 Extending an emulation platform for automatized and distributed evaluation of QoS in MANETs Wannes Vossen Alvaro Torres Jorge Hortelano Carlos T Calafate Juan Carlos Cano and Pietro Manzoni Department of Computer Engineering Universidad Polit cnica de Valencia Camino de Vera S N 46022 Valencia Spain wanvos posgrado upv es a
28. 3 Even when two nodes remained out of range during the whole simulation they had some data transferred during the simulation This problem came from an error with the synchronization between client startup and visibility settings With static simulations it was not a problem to set visibility instructions 10 seconds before the simulation started those instructions did not change during the whole simulation On the contrary it allowed to spread the Access Points system load on simulation startup With mobility visibility instructions are dynamic and change over time Those 10 seconds of delay introduced incoherent simulation results visibility settings were always 10 seconds earlier than traffic instructions 4 totalTime 111 was hard coded somewhere in the code This caused an out of array exception if the simulation lasted less than 111 seconds In order to validate the Mobility Plugin System I generated a mobility scenario that makes nodes move in diagonal with different speeds for each node The JAVA code for that algorithm can be found in Appendix H on page 85 I then manually calculated the theoretical position of a few nodes at a determined time and checked it was well imported in Castadiva This is probably not a full validation of the Mobility Plugin System and it would probably be interesting to compare results But I did not succeed in obtaining any valid results with mobility in Castadiva Important note The Mobility Plugin syst
29. 512 1000000 Pkts Second Max Packets Throughput Kb s Pkts Received 400 0 0 10 0 0 0 110 AP APL UDP 512 400 0 0 0 0 EI o 10 110 UDP KET TEK 512 400 0 0 0 0 Duplicate Row Delete row Status waiting Order traffic Delete all UDP Packets Reveived 0 Save in text file Average Throguhput 0 Kb s Close Figure 4 Traffic window 4 2 Delay measurement for UDP traffic Measuring the delay of UDP packets is a prob lem which is not easy to solve in a real environ ment Notice that in a simulation environment this is very easy to do as we can determine when an UDP packet leaves the source and when it arrives to the destination However in a real environment we have to deal with clock desynchronization which significantly compli cates the process To cope with this problem we implemented two different methods one more precise and another just orientative The basis of both methods is our UDP flow generator At the source node we read the local time and stamp it on each sent packet On the sink applica tion again we obtain the time and then com pare it with the packet s time stamp The first method counters clock desynchro nization by relying on ntp 1 at the beginning of each emulation This method is just orienta tive since it is not able to solve the clock drift problem during tests
30. 62 432094828450175 node 2 setdest 1240 0 560 0 59 38569098715311 Node 2 NEW TARGET 1240 0 140 0 Node 2 TARGET REACHED 1240 0 140 0 58 1799813877065 Km h ns at 88 43832044864641 node 2 setdest 1240 0 140 0 58 1799813877065 Node 2 NEW TARGET 60 0 140 0 ns_ at 120 0 node 2 setdest 718 4845036632628 140 0 59 5073149544258 NODE 3 MOVEMENTS Node 3 TARGET REACHED 500 0 380 0 67 02568486533434 Km h ns_ at 12 611064243209228 node 3 setdest 500 0 380 0 67 02568486533434 Node 3 NEW TARGET 1400 0 380 0 Node 3 TARGET REACHED 1400 0 380 0 84 05963188698215 Km h ns_ at 51 19406600728597 node 3 setdest 1400 0 380 0 84 05963188698215 Node 3 RED SEMAPHORE UNTIL 66 41438630266265 Node 3 GREEN SEMAPHORE Node 3 NEW TARGET 1400 0 1440 0 ns_ at 66 45 node 3 setdest 1400 0 380 0 0 0 Node 3 SPEED CHANGE 0 0 Km h gt 77 55136935175538 Km h Node 3 TARGET REACHED 1400 0 1440 0 77 55136935175538 Km h 75 ns_ at 115 65609438489076 node 3 setdest 1400 0 1440 0 77 55136935175538 Node 3 RED SEMAPHORE UNTIL 129 19274902868204 ns_ at 120 0 node 3 setdest 1400 0 1440 0 0 0 NODE 4 MOVEMENTS Node 4 TARGET REACHED 800 0 340 0 59 76214259985875 Km h ns_ at 21 281796117941216 node 4 setdest 800 0 340 0 59 76214259985875 Node 4 NEW TARGET 1440 0 340 0 Node 4 TARGET REACHED 1440 0 340 0 92 42208279251665 Km h ns_ at
31. 818 Phy WirelessPhy set freq _ 914e 6 Phy WirelessPhy set L_ 1 0 Phy WirelessPhy set bandwidth 11e6 Phy WirelessPhy set bandwidth 54e6 Initialize the 802 11 MAC Mac set bandwidth_ val bw puts Bandwidth set to val bw Mac 802 11 set dataRate val bw Configuration for 802 11a and 802 11g Mac 802 11 set CWMin_ 15 Mac 802 11 set CWMax_ 1023 Mac 802 11 set SlotTime 0 000009 Mac 802 11 set SIFS_ 0 000016 Mac 802 11 set PreambleLength _ 96 Mac 802 11 set PLCPHeaderLength_ 40 Mac 802 11 set PLCPDataRate 6 0e6 Mac 802 11 set RTSThreshold 3000 Mac 802 11 set ShortRetryLimit_ Mac 802 11 set LongRetryLimit_ Mac 802 11 set basicRate 6e6 Mac 802 11 set dataRate 54e6 AN for 802 11a and 802 11g 20us 10us 144 bit 48 bits 1Mbps bytes retransmittions retransmissions 73 D CityMob2 output file example WO J O U1 E UKM KA MODEL 4 Enhanced Downtown Traffic Simulation Model SEED 1273842116345 nodes number 5 max time 120 0 max speed 100 0 max X 1500 0 max Y 1500 0 d 20 0 damaged 1 alpha 0 5 delta 5 0 min speed d 25 0 max speed d 50 0 HHH HHH HHH HH HH Node 1140 0 1390 6241869639089 gt 1140 0 280 0 58 88932070560723 Km h node_ 0 set X 1140 0 node 0 set Y 1390 6241869639089 node_ 0 set Z 0 0 Node 1 540 0 111 77523584641058 gt 540 0 1200 0 75 06782928526441 Km h node 1 set X 540 0 node 1 set Y 111 77523584641058
32. Additionally the low margin of error required can not be achieved with ntp The second method we adopted was the so lution proposed in 2 It employs the control network Fast Ethernet based used by Cas tadiva to redirect UDP packets back to the source when using this method both source and sink programs are on the same machine This way we avoid clock desynchronization but we introduce a new delay This delay should be measured and taken into account when results are processed In terms of implementation and integra tion with Castadiva we can select the second method indepently for each flow by setting the Redirect flag to true as we can see in fig ure 5 4 3 Plugins system The first version of Castadiva was fully func tional but only allowed using a single mobility model and three routing protocols With the dynamic creation and loading of plugins we have managed to support almost any routing protocol and mobility model Routing protocols Previous versions of Castadiva already sup ported a few common routing protocols For example OLSR 9 and AODV 3 For those protocols to be functional they had to be previously installed and configured on each node What Castadiva did when a simulation started was to activate the selected protocol sending a simple command line in struction via Secure Shell SSH When the simulation ended it was deactivated the same way However a few constraints come up when
33. DIVADIR castadiva NFSDI GW 1 ROUTE R castadiva nfs 92 168 1 2 R 201 Creation of the castadiva directories mkdir chmod mkdir chmod CASTADIVADIR 0777 CASTADIVADIR NFSDIR 0777 NFSDIR Configuration of temporary internet access for package install route route echo del default add default gw GW nameserver 158 42 249 8 gt etc resolv conf NFS install opkg opkg update install kmod fs nfs 23 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 Set automatic NFS mount at boot echo mount tunfsu GW CASTADIVA NFSDIR ounolock gt gt etc rc d S90custom user startup sh chmod 0777 etc rc d S90custom user startup sh SSH Public Key configuration cat id_ pub gt gt etc dropbear authorized_keys chmod 0600 etc dropbear authorized_keys Network configuration cp network etc config network cp system etc config system cp wireless etc config wireless cp firewall etc config firewall echo Install successfull please reset router The script and its attached files can be copied on the router using the scp r command and it should be run on the router using ssh Let us now review the different files and commands that are used in this script e As we configured a default gateway NAT and routing should be enabled on the gateway so that an Internet connection can be established The DNS value mus point to a valid DNS server echo 1 gt
34. DOM WAY POINT x L Range M Labels 1990 O Graph C Movement Routing protocols olsr x 00m E n New Simulation Replay Simulation TET Stop Simulation Show Save Scenario Random Scenario Drawing area 1500 Nod s Tan ETO oN ONE NRA Ae Te Reset all AP 4308 500 1500 2500 3500 4500 5500 6500 7500 8500 500 m Help Traffic SS Declare traffic Status Ready gosa Figure 4 3 Castadiva s simulation window Figure 4 3 shows the graphical interface that was designed to easily control the previously described system As we can see it is possible to place each Node on a canvas and to define its range According to those ranges Castadiva makes sure that two out of range Nodes do not communicate even if their real interfaces are able to do so Among the options available in figure 4 3 we can see that some are related to Mobility and Routing Protocols Those features are part of my work on Castadiva and will be described in section 6 2 on page 25 P CASTADIVA TRAFFIC all Traffic Control Start s Stop s Source Address Traffic Transfer Size B _ Size of packet Pkts Second Max Packets Throughput Kb s Pkts Received 69 10 110 AP 1 AP2 UDP 512 4 400 0 0 0 0 110 AP3 1000000 0 0 1110 AP4 512 4 400 0 0 0 0 110 512 4 400 0 0 0 0 Duplicate Row Delete row UDP Packets Reveived 0 gt Save in text file Status waiting Order traffic De
35. GUT Let us review each option offered by figure 6 7 The Name field allows us to identify the protocol and it s configuration It must be unique and if an existing name is re used it will be overwritten Bin must point to the location of the binary file on the Access Point which allows to start the routing protocol Note that the whole path must be given and that the last part of that path which should be the filename is also used to kill the protocol when the simulation ends This is done trough the killall binName command Therefore flags will be inserted just after Bin and for protocol startup only Configuration file content is the content of the configuration file that will be written in the Path protocol conf on each Access Point Important note The automated compilation uses certain files in the source src folder of the project Make sure that the following constants are correct in Castadiva Model java e PLUGIN_WORKFOLDER pluginTemporaryFiles is the name of the temporary folder that is created for compilation purpose 31 32 Configure Protocols New Protocol Configuration Bin Jusr sbin olsrd Flags Configuration File Content LinkQualityAlgorithm etx_fpm Interface wlo Ip4Broadcast 255 255 255 255 Hellointerval 2 0 HellovalidityTime 40 0 Telnterval 5 0 TcValidityTime 100 0 Midinterval 18 0 MidValidityTime 324 0 Hnalnterval 18 0 HnaValidityTime 108 0 P
36. Internship in the Universidad Polit cnica de Valencia Validation and improvement of a MANET emulation tool Wannes VOSSEN wanvos posgrado upv es June 22 2010 English The present document is a detailed report of a five months internship in the computer networks research group of the UPV in Valencia Spain The described work placement is part of the first year of a Master degree in computer networks and telecommunications Master STRI Toulouse France In this report I will focus on two aspects I relate the experience of such an international internship and II present the results and conclusions of the achieved work Readers might be interested in only one of those two aspects and should consider the table of contents to sort relevant information Espanol Este documento es la memoria de una practica de cinco meses en el grupo de investigaci n de redes de la UPV Valencia Esta practica es parte de un primero ano de Master en redes y telecomunicaciones Master STRI Toulouse Francia En la memoria describir dos aspectos en particular I Intentar transcribir la experi encia de tal instancia al extranjero y II presentar los resultados y conclusiones de mi trabajo El lector pueda tener inter s en uno solo de estos dos aspectos y puede referirse al index para encontrar su contenido Francais Ce document est un rapport d taillant cing mois de stage dans le groupe de recherche sur les r seaux informatiques U U
37. PV Valencia Espagne Le dit stage a t effectu dans le cadre d une premi re ann e de master en r seaux et t l com munications Master STRI a Toulouse Ce rapport ce concentrera principalement sur deux aspects I il s attachera retranscrire l exp rience d un stage l tranger et II il pr sentera le travail accompli et les r sultats obtenus durant le stage Le lecteur pourrait n tre int ress que par un seul de ces deux aspects et devrait se r f rer la table des mati res pour trouver l information souhait e 1Grupo de Redes de Computadores See http www gre upv es Universidad Polit cnica de Valencia see http www upv es 3Syst mes de T l communicatons et R seaux Informatiques See http www stri net Contents Personal Experience 7 Background and motivations 8 The Host University 9 2 1 Universidad Polit cnica de Valencia UPV 9 2 2 Grupo de redes de computadores 9 2 3 An internship opportunity 10 2 4 Technical aspects 22 4 43 48 ee du a he da ds 10 Objectives amp time management 12 Belts ODICCH VES axetil tans Z a eet Sante shied kin GMa ASE Oe le 19 12 3 2 Time management sore e a ea kek ee ee ee oe do hae a 12 Achieved work 14 What is Castadiva 15 A MANES nadie ne re Meg RL Reus inner HT 15 4 2 Introduction to Castadiva 15 4 3 Castadiva s features
38. ac Mac 802 11 MAC type set val ifq Queue DropTail PriQueue interface queue type Queue DropTail PriQueue for AODV CMUPriQueue for DSR 1 set val ifq CMUPriQueue for DSR set val tl LL link layer type set val ant Antenna OmniAntenna antenna model set val x 2000 X dimension of the topography set val y 2000 Y dimension of the topography set val ifqlen 50 max packet in ifq set val seed 0 set val rp AODV one of OLSR AODV AODVUU DSR TORA OLSR v7 set lon 200 how many nodes are simulated set val nn_traffic_sources 50 nodes que navegaran por web set val nn servers 1 nodes servidores set val cp traffic trace n50 c5 file patron de conexiones en web no utilizar set val sc scenarios m2 n200 t6000 s17 w2000 h2000 d80 a0 vl scn escenario de trafico CAMBIAR set val stopns 20 simulation time set val trfile out tr set val namfile out nam set val httpfile out http set val random mobility FALSE set val random number generator TRUE set val bw 54e6 set val completion 0 set val nn_pages 300 set val nn_session 100 interSession 1 interPageOption 0 pagesize 10 objsize 10 set val set val set val set val at aii PS i Sen check for boundary parameters and random seed if val x 0 val y puts gt No X Y boundary values given for wireless topology n if val random number generator TRUE puts g
39. aro ScenTODO FG home alvaro ScenTODO FH 10 Ready BCE New Simulation Load Scenario home alvaro ScenTODO GH home alvaro ScenTODO GH 10 Ready Edit propierties Import NS 2 Scenario Import CityMob Scenario Repetitions 17 S Accept Source Folder home alvaro ScenTODO FG Search Results Folder _ home alvaro ScenTODO FG Search Edit Simulation Delete Simulation Load list Save list Generate Simulations Close Figure 8 Execution planner GUI ferent simulations that Castadiva will process column Source folder and it also includes the path where results will be saved Results folder the numbers of runs per simulation Runs column and at the end it includes a message that informs the user about the status of the simulation Ready Simulation in progress Canceled Retrieving results or Done The results are stored in a directory named Iterations inside the folder we mentioned in the column Results Folder Inside that di rectory every simulation creates a file named X _DefinedTraffic txt where Xe 1 Runs where each file contains the data of the table that we can see in figures 5 and 4 An example of this file can be seen in table 1 5 Conclusions and future work In this work we present some extensions to the Castadiva platform to improve research in the MANETs field by allowing to make real test bed exper
40. art from those systems I had to validate I also had to perform editorial work e Participate the writing of a research article dealing with Castadiva e Write a report and documentation dealing with the knowledge I gathered along the internship Of course these are the main tasks and they oriented my work They led me to achieve other subsequent tasks like e Setup of the test bed e Understanding of the existing program e Learning how to use simulation tools in ordre to validate Castadiva s results In the next chapter each of those tasks is described with more detail 20 6 Results 6 1 How to setup the test bed The following will explain how to setup a computer as a Castadiva Core and five LinkSys WRT54GL routers as its nodes In figure 6 1 we can see the exact architecture I used for my work it is also the architecture that I will use for this example In figure 2 2 on page 11 we can also see a picture of the real test bed yy 201 D 2 yy tas POT yy 204 yy 205 Ethernet 192 168 1 0 24 Castadiva Core Figure 6 1 Possible architecture for the Castadiva Test Bed It is important to note that this example was tested with certain software versions Commands filenames might change with newer versions Nevertheless the following in structions can be extrapolated for a general use including other types of node devices 6 1 1 Setup the Core To implement the core
41. ath protocol conf Jetc olsrd conf Create plugin Figure 6 7 Routing plugin configuration window e PLUGIN_JAR_FOLDER src castadiva Plugins is the path to the Plugin folder where jar files are copied after compilation and where Plugins are loaded on startup e PLUGIN_INCLUDE_FOLDER src lib must point to the folder containing the IPluginCastadiva java abstract class 6 2 2 6 Troubleshooting The following is presented as a Question Answer list 1 J OO G LA A I just installed or used a custom routing protocol on my Access Points Now I can no longer execute a simulation with the NONE routing set tings Two things should be checked First the routing protocol should not start auto matically when your Access Point starts See section 6 2 2 5 to disable the routing protocol on startup Secondly try to run the route command on your access points A previous simulation may have stopped unexpectedly and did not stop the rout ing protocol Use the killall protocolBinFileName command to make sure that the protocol is no longer running After I tried to configure a plugin Castadiva wont launch anymore Try removing the jar file from the plugin directory A folder is created in Castadiva s execution directory What about it A folder usually named pluginTemporaryFiles is created to compile and then package the plugin This folder should be deleted automatically and it can be har
42. ation For example professors are in charge of PhD students who are working on their thesis and participating the research work Professors can also appeal to Master students as a sup port for their research It is a possibility for Spanish Master students to work in their university s laboratories and receive specific grants in return Master students can also be foreign students who have to perform an internship during their exchange This is my situation For my every days work in the laboratory I collaborated with Jorge Hortelano and Alvaro Torres Jorge is a PhD student and he is also the creator 1 of the software I worked on He introduced me to its functioning and answered many of my questions Alvaro is a Master student he also worked on the program I had to validate We thus had to work together as one s changes could interfere with another s one As he had been working on the software for a longer time he also answered many of my questions Alvaro and I also collaborated to write the research article described in chapter 6 5 on page 49 Faculty Raquel Lacuesta Miguel S nchez Alberto Gonzales Miguel Mateo Pablo Boronat Lourdes Pe alver Pietro Manzoni Coordinator Carlos Miguel Calafate Juan Carlos Cano N Carlos Lino Jorge Hortelano Francisco Martinez Dominguez Ingrid Juliana Ni o Gonzalez Margarita Nacher Jorge Belenguer Johann M Marquez Barja Master Students PhD Students Adrian Puchades
43. diva and 8 2 yep ana eis y a ess Le ae nn ee de et SS 52 hide GED 0 OT aana a e Pie ao ie A Ge ve a Se Sh tec a ad 52 eS ATER RE at cline AE E E aA Seta eames ete od ke io 53 8 Conclusion 54 9 Thanks 55 Bibliography 57 III Appendices 58 A Castadiva s GUI splash view 59 B Extending an emulation platform for automatized and distributed evaluation of QoS in MANETs 2 61 C Johann M Marquez Barja s implementation for 802 11g in ns 2 70 D CityMob2 output file example 74 E Configuration for 802 11g in ns 2 without distance losses 77 F Configuration for 802 11g in ns 2 with distance losses 80 G Simple Gnuplot example file 83 H Mobility Plugin algorithm for the Mobility Plugin System 85 List 2 1 2 2 3 1 4 1 4 2 4 3 4 4 6 1 6 2 6 3 6 4 6 5 6 6 6 7 6 8 6 9 6 10 6 11 6 12 6 13 6 14 6 15 6 16 6 17 6 18 6 19 of Figures Organization chart of the GRC research group Working material with the stacked routers Task distribution over time Castadiva s architecture Software components for Castadiva Castadiva s simulation window Castadiva s traffic window Possible architecture for the Castadiva Test Bed Global design of the Execution Planner 00004 Simulation edition i
44. e Load Scenario a X tesa Wifi IP Address NFS Directory Wif IP Address NFS Directory m Save cults i testay zoom Position ran bete tet 192 168 2 201 icastadivenfs 192 1682 201 icastadivanfs x 619 0 m m R as Mode Device channel Mode Device channel keat eed retd lt Y 437 0 m Exit ct Ns2 D TETT Ad Hoc x wo IE E Ad Hoc s wo B 205 routes 2m m T Wifi Mac Address Processor Wifi Mac address Processer Time G Ms 00 LD 7E 30 68 63 mes ly 00 1D 7E 30 9 63 mes iter EE LZ Simulation putes E ITFTes w Gateway Gateway otal ite 7 192 168 12 Forward Bee a LK MF connection S D S r range Baz m Configuration Help Node Mobility Generate Simuletion bre SSH User SSH Password User Password G G Rendom Simulation root em n Max speed ED s g I 600 rn Execution Planner E pase 00 s S F Add Hel change Moby Model Lee View word ve mom Les close myPlugin 0 Range WI Labels et lt T l sa O Graph C Movement Routing protocols Z L ir tom marnas E et de Kaa es AP net devices B E r Ethernet device wifi device Replay Simulation Z as Ailai external applical Eros 1400 m Stop Simulation Configurate the Compyt BRIE oe Switch device Bridge device ody an AP fer I Slam Tas T Gr x S Random scenario Sais en Protocol udp Net addresses f rawing araa 500 Load AP list ie en me E TS Ethernet iP vpr a a mi LT ane fon H Node pes aE PRT eset al esas fi92
45. e I believe I made an important progress and I probably identified the source of most of the remaining errors I hope someone will continue my work and therefore I wrote this report with as much detail I could My experience as an trainee in the GRC group and in the UPV in general is defini tively positive Even though I had some hard times desperately looking for an error I enjoyed most of the hours I spent in the laboratory The people I met there are nice friendly and ready to help This is probably also an important aspect when it comes to one s professional life Regarding knowledge and competences I probably augmented my analysis abilities as well as the ability to progressively handle a large problem Of course I dealt a lot with JAVA Linux and ns 2 but I also increased my general computer network knowledge as well as my command of the Spanish language 54 0 Thanks There are a few persons I would like to thank for their help with this internship project Pietro Manzoni Coordinator of the GRC For being my internship director and for of fering me a job in the Computer Network Lab at the UPV Andr Aoun Director of STRI For his voluntary support all along my project from it s very beginning to it s last days Marta Caballero Exchange Coordinator at the UPV For her patience an her support with administrative tasks Carlos Miguel Calafate Professor at the UPV for his pertinent technical advice Alvaro Torres
46. e is added to the execution planner To edit the line and change the source folder the results folder or the amount of runs use the Edit Simulation button See figure 6 3 26 P CASTADIVA Execution Planner Jox Source folder Runs T Status 10 Ready Results folder home wannes Desktop ns2 Si home wannes Desktop ns2 Si New Simulation Load Scenario BEE CE i Source Folder tymobycastadivaThroughputUDp Search Edit Simulation Results Folder tymobycastadiva ThroughputUDp Search Delete Simulation e La Repetitions 9 d Accept Cancel Load list Save list O Start simulation at Hours o 5 Minutes bo RDJ Generate Simulations Close Figure 6 3 Simulation edition in the execution planner The Source folder is the path to your scenario and the Result folder will contain the simulation results The results are written in a text file named X_definedTraffic tat where X corresponds to the run For example if you have 10 runs in your simulation the first run will be written in the Result Folder Iterations 10_definedTraffic tat and the last run to Result Folder Iterations 1_defined Traffic tat The Load list and Save list buttons allow to save and load the execution planer s simu lation list to a file It might be interesting to save a large list in order to save time in the case som
47. e of Castadiva Later on Alvaro and I investigated the possible reasons of such a speed difference We finally found out that there probably was an error with the OpenWRT router s drivers and UDP Let us now describe the different tests that led us to such a conclusion 1 Our first reaction to that low throughput was interference probably due to the The delay is depending on the ARP_PERSISTENCE parameter The default delay is 31 seconds See http www rabbit com documentation docs manuals tcpip usersmanualvi arp_dns html 47 Throughput Kbits second 48 802 11g throughput depending on packet rate in Castadiva and NS Distance between nodes is 5cm transport protocol is UDP 8000 Castadiva simulation 1 7000 Castadiva simulation 2 Castadiva simulation 3 6000 NS Simulation 5000 4000 3000 2000 1000 0 0 500 1000 1500 2000 2500 Packet rate packets second One packet 512 KBytes Figure 6 18 UDP CBR Throughput comparison between ns 2 and Castadiva election of a bad channel But TCP transfers do reach much higher speeds we reached up to 8Mbps with TCP instead of UDP In Alvaro s configuration with EEEPCs instead of routers everything is fine We then tried to connect the routers with Ethernet and ran the same UDP simula tion We reached around 25Mbps which is acceptable In order to know if the problem came from the UDP client software we used an EEEPC as a client to generate the packets
48. e section 4 2 to learn more about the Core in Castadiva 2 What You See Is What You Get 50 7 2 2 Learning about ns 2 There are many tutorials dealing with any aspect of ns 2 spread all over the Internet For example I recommend Marc Greis s tutorial which offers an introduction to ns 2 s basics and also teaches the basics of wireless simulation with ns 2 In general I would recommend to focus on this page to encounter information about ns 2 As there are several versions of ns 2 I rarely succeeded in finding appropriate information using any search engine I thus recommend to use as much as possible the NS Manual in order to find precise and probably version related information 7 2 3 Simulating 802 11g with ns 2 In order to validate results that were generated using 802 11g routers a first step is to configure ns 2 with the correct parameters to simulate 802 11g Surprisingly I did not find any set of parameters to achieve this on the Internet I finally investigated two different options to solve this problem 1 Johann M M rquez Barja also working in the network laboratory implemented 802 11g for his own work See appendix C for more detail I used his implementation as a basis of my own configuration file that does not support distance losses and thus closer to Castadiva See appendix E on page 77 for the complete configuration file 2 Later on a new version of ns 2 ns 2 3 was released One of the important im
49. em is functional but the results of any subse quent simulation can not be considered as valid This is because of a problem that was discovered in the Access Point s client and server programs in particular with TCP This problem is affecting all mobility in Castadiva even random mobility The issue was not fixed but it is detailed in section 6 3 1 on page 44 38 x B Gitymob 2 0 Help Simulation Parameters Nodes 50 MaxSpeed 100 Max X 100 Min Speed 50 Max Y 100 Max Speed POI 50 Max Time 60 Min Speed POI 25 Street Dist 20 Alpha 0 5 Lanes 1 Delta 5 Accidents 5 Semaphore Prob 0 3 Semaphore Pause 20 Points of Interest Add Remove Output File Start Simulation Figure 6 10 CityMob 2 0 User Interface 6 2 4 CityMob Import 6 2 4 1 Introduction CityMob See figure 6 10 is a mobility pattern generator for VANETs Vehicular Ad hoc NETworks It allows to generate random mobility scenarios in ns 2 format An example of CityMob output can be found in appendix D on page 74 Next sections will test fix explain and validate CityMob importation in Castadiva 6 2 4 2 Design This section shows the current design of the CityMob importation system The global behavior and design respects Ignacio Munoz Vera s initial work but the implementation of each function suffered many changes as it corrects a very important semantic errors as well as a syntact
50. en though the previous test succeeded the validation of the Routing Protocol Plugin is not achieved it would probably be interesting to try to implement another protocol I was suggested to implement AODV but I did not succeed as there is no official implementation for AODV in OpenWRT Kamikaze Let us finally comment the design of the Routing Plugin System Using the ServiceLoader class to simply store a few text lines is probably a very complicated way to enable Routing Plugins in Castadiva A simple text file would have been an easier solution Moreover it would be safer as it would not require to compile and package the plugin files Calling system commands might lead to cross platform compatibility issues Also to make it more flexible the routing plugin system could be replaced by a larger system that would simply allow the execution of custom user code before and after a simulation 6 2 3 Mobility Plugins 6 2 3 1 Introduction I previously commented the plugin concept in section 6 2 2 on page 28 A Mobility Plugin is supposed to enable custom mobility pattern definition in Castadiva In other words Mobility Plugins can be configured to add controlled and dynamic mobility to Castadiva s scenarios In previous versions Random Mobility already was an option as well as manual mobility which was not handled by Castadiva Controlled mobility allows the usage of an algorithm to define the positions of the nodes during the simulation The Mobi
51. es One node s X position in a random simulation depending on simulation progress 1400 1200 1000 800 600 400 200 0 20 40 60 80 100 120 140 Simulation progress Seconds NS node 1 NS node 3 NS node 5 Castadiva node 1 Castadiva node 3 Castadiva node 5 NS node 2 NS node 4 Castadiva node 2 Castadiva node 4 Figure 6 14 One node s X position in a random simulation One node s Y position in a random simulation depending on simulation progress 0 20 40 60 80 100 120 140 Simulation progress Seconds NS node 1 NS node 3 NS node 5 Castadiva node 1 Castadiva node 3 Castadiva node 5 NS node 2 NS node 4 Castadiva node 2 Castadiva node 4 Figure 6 15 One node s Y position in a random simulation 43 6 2 5 Minor issues in Castadiva This section contains a list of issues I encountered in Castadiva Those issues are minor issues that are not directly related with the parts of the program I worked on I fixed some of those issues 6 2 5 1 but had no time to fix all of them 6 2 5 2 6 2 5 1 Fixed issues 1 When the Execution Planner the Routing protocol designer the Simulation or the Mobility plugin designer windows where closed using the upper right cross the whole program was exited with no warning 2 The pa
52. ess e Support routing route command for static routing e Support iptables They are used to drop packets when nodes are logically out of range but fiscally in range Which typically occurs with stacked nodes In our example we use LinkSys WRT54GL wireless routers with an Open WRT Linux Based operating system OpenWRT is a Linux distribution for embedded devices It is a full operating sys tem based on a Linux kernel that supports several packages and offers an important customization level The operating system is supplied as a pre compiled firmware that can be sent to the device using the original manufacturer s firmware update system A detailed and up to date tutorial of how to install OpenWRT is available at http wiki openwrt org oldwiki openwrtdocs hardware linksys wrt54gl The follow ing is a commented summary of the tftp method 1 Warning You MUST install the 2 4 kernel before you install any more recent kernel Otherwise you wont be able to flash using tftp anymore 1A Linux Based router that costs approximately 50 See http wuw Linksysbycisco com EU en products WRT54GL See http openwrt org 22 GC I Oo e MA DH 7 Download the desired version of the firmware on the OpenWRT website It should come as a bin file For example openwrt wrt54g squashfs bin Install a Trivial File Transfer Protocol TFTP client on the computer For example atftp is a good option for Debian
53. ething goes wrong during simulation The Start simulation at checkbox offers to defer the execution according to the system time the execution planner will figure out how long it must wait before the configured time of the day Hours and Minutes The simulation can thus be planned up to 24 hours after it s configuration Note that the real start time can be delayed of up to one minute and that the timer can be canceled as long as the simulations are not started Finally the Generate Simulations button starts the execution planer s process During simulation the runs will be decremented and the Status information for each row will be updated The status can take one of the following values e Ready The scenario is ready to be simulated e Done when the simulations where successfully played for this scenario e Waiting for the AP when Castadiva is collecting simulation information It means that the simulation is ending If the Execution Planner gets blocked on this message 27 it might mean that something went wrong during simulation and that it is not able to collect the simulation information e Simulation in process the scenario is being simulated e Canceled the scenario s simulation was interrupted When simulating the list the Reset Access Points is enabled Clicking this button simply sends a reboot instruction to every access point This might be a solution to unlock a problematic simulation or cancel the execut
54. g a sleep command 8 The routing instructions were sent too early The routing protocol was started seven seconds before the client and visibility settings This was no problem with static routing as routes did not change along the simulation Moreover separating those instructions was interesting to reduce the Access Point s load But with dynamic routing Plugins the routing protocol had 7 seconds to initialize itself with no visi bility restrictions As the plugin system should be generic for any routing protocol it is not possible to evaluate the initialization time The initialization time should thus be part of the simulation I made a few simple tests in order to check the creation of a plugin and the loading un loading of a Routing Plugin during a simulation Therefore I aligned 5 nodes so that each node only saw his direct neighbor I then connected to every Access Point using SSH and checked that the routes were set correctly using the route command Finally i connected to the lefter Access Point and ran the traceroute command to the righter one The results are correct root router1l traceroute 192 168 2 205 traceroute to 192 168 2 205 192 168 2 205 30 hops max 38 byte packets 34 1 192 168 2 202 192 168 2 202 2 370 ms 2 953 ms 8 983 ms 2 192 168 2 203 192 168 2 203 5 417 ms 4 306 ms 5 376 ms 3 192 168 2 204 192 168 2 204 6 888 ms 6 809 ms 6 434 ms 4 192 168 2 205 192 168 2 205 8 055 ms 9 685 ms 7 111 ms Ev
55. g on simulation progress 1400 a 1200 g Z 1000 2 9 800 oO x OD 600 5 o 400 zZ 200 0 E 0 20 40 60 80 100 120 140 Simulation progress Seconds NS node 1 NS node 3 NS node 5 Castadiva node 1 Castadiva node 3 Castadiva node 5 NS node 2 NS node 4 Castadiva node 2 _ Castadiva node 4 Figure 6 12 One node s X position in a random simulation without error correction One node s Y position in a random simulation depending on simulation progress Q oO t c S Q oO gt Nn o TD Z 0 20 40 60 80 100 120 140 Simulation progress Seconds NS node 1 NS node 3 NS node 5 Castadiva node 1 Castadiva node 3 Castadiva node 5 NS node 2 NS node 4 Castadiva node 2 Castadiva node 4 Figure 6 13 One node s Y position in a random simulation without error correction Al condition if we are processing the last move of the node which would be the last second of his movement then we set the node at his theoretical destination which is given by the movement instruction Results for the same simulation can be found in figures 6 14 and 6 15 on the next page Important note The CityMob user interface show speeds as Km h In the generated ns 2 files comments for each movement instruction also announce Km h The same values are inser
56. ic errors Figure Figure 6 11 on page 40 shows how Castadiva deals with CityMob files city Mob out and converts them into a set of Access Points and a mobility matrix which is a matrix that contains the position of each node at every second of the simulation When a CityMob importation is processed Access Points parameters like its name its IP address can come from two possible sources 1 A first and priority option is the aps tzt file which can be configured in the configu ration folder of Castadiva That file contain settings for a list of node The first line will be used for the first node and so on Note that aps tat is similarly used with the node creation utility See http www grc upv es Software citymob html See 6 2 4 4 for more details about the errors 39 ES L L S numberOfNodes pause maxSpeed OR Castadiva default values aps txt l Mobility Instructions ReadNSNode ReadNSMobility best rev e ee oles Castadiva Access Point translateNSDataToMobilityMatrix creation with initial position J Castadiva Mobility Matrix sr Figure 6 11 CityMob importation process 2 second option is to use Castadiva built in default values they are defined in CastadivaModel java s header Let us now comment the most important function in figure Figure 6 11 on page 40 translateNSDataToMobilityMatrix That function takes a list of movement instructions for examp
57. il able without any previous infrastructure like emergency missions military operations or ad hoc meetings Testing and evaluating protocols for MANETs is a mandatory request to guar antee its success in a real world application Researchers in this field have three options for testing their MANET protocols using simulation tools using emulators or using test beds Emulators provide an attractive middle ground between pure simulators and wireless test beds allowing scalable and repeatable ex perimentation using real devices In section 2 we review the different emulators available Castadiva 8 is one of those emulators it is also the platform that we are extending in this paper Further details can be found in section 3 Also the proliferation of devices with multi media and wireless networking capabilities in crease the demand of audiovisual communica tions anong MANET nodes requiring certain levels of QoS To meet this need the IEEE 802 11e 7 working group has enhanced the IEEE 802 11 6 MAC to provide QoS at the MAC layer So in order to meet these new re quirements we have updated Castadiva to be QoS capable The rest of this work is organized as fol lows a few related works are presented in sec tion 2 Section 3 briefly describes the original Castadiva architecture and features In sec tion 4 we present the different extensions made to Castadiva to offer QoS support Finally in section 5 we present s
58. ime management is that I have attended classes during the entire internship I attended 9 hours of class every week homework and projects represented at least another 9 hours of personal work This taken in account the instructions I received from my home university were to balance as good as possible among those two activities Pietro Manzoni also warned me that he would not be watching at my timetable I thus had to discipline myself trying to be as efficient as possible in my university work in order to be able to spend as much time as possible in the laboratory I also spent fewer hours in the laboratory during the exam period which was the week before the 11 and 15th of June Every month on Wednesday the whole group meets and discusses the student s works Unfortunately I had a class on Wednesday overlapping with the meeting I gave up one of those classes to participate one of those meetings I think it is an important experience to understand how the laboratory works When I presented my achieved work I was pleased to hear critics and interesting suggestions to deal with my problems 12 01 0 08 0 15 0 22 0 01 0 08 0 15 0 22 0 29 0 05 0 12 0 19 0 26 0 03 0 10 0 17 0 24 0 31 0 07 0 14 0 21 0 2 10 2 10 2 10 2 10 3 10 3 10 3 10 3 10 3 10 4 10 4 10 4 10 4 10 5 10 5 10 5 10 5 10 5 10 6 10 6 10 6 10 ask Week et up the test bed and the work environment tudy of the NS 2 simulator for alidation purposes 3 4 5 6 7 8 9
59. iments in a simple and straightfor ward manner With the proposed improvements we have created a simple and powerful tool to automa tize the evaluation of MANETs in general and QoS enabled MANETs in particular Now the VoIP and Videoconferencing experiments can be done taking into account QoS and delay pa rameters which are usually available for sim ulation studies but only in a very few real ex periments With this new features we plan to create a real test bed to analyze the behavior of vari ous routing protocols when their packets are prioritized via IEEE 802 11e Castadiva is free software developed under the GNU GPL license and can be downloaded at http castadiva sourceforge net Acknowledgments This work was partially supported by the Min isterio de Educaci n y Ciencia Spain under Grant TIN2008 06441 C02 01 by the Gener alitat Valenciana under Grant GV 2009 010 Strt Stop Src Addr Traff Transf P sec Thrgpt Received Delay AC Redir 10 130 6 3 TCP 10737 5823 931 10 130 6 3 UDP 512 125 505 941 99 64 121 64 VI true 10 130 6 3 UDP 256 10 20 277 100 0 87 209 VO true 10 130 3 6 UDP 512 125 504 497 99 52 84 506 VI true 10 130 3 6 UDP 256 10 20 156 99 4 53 654 VO true Total UDP packets received 99 64 Average throughput 1374 9603 Table 1 Sample results obtained by the execution planner References and Information exchange between sys
60. in System but they were designed together and do therefore share a similar design In the validation section I will make suggestions for an alternative design The Java programming language offers a mechanism to allow a user to insert custom code directly into a compiled program More generally the term of Plugin can be used This is a non exhaustive list for the Kamikaze version of Open WRT It is the result of the opkg list grep routing command See http www olsr org for further information 29 v B olsrdijar s A META INF s services 4 libiPluginCastadiva S MANIFEST MF s castadiva s Plugins olsrd class s lib S IPluginCastadiva class olsrd conf Figure 6 5 Structure of a routing plugin s jar file As shown in figure 6 4 the ServiceLoader class can load a jar archive directly into the a Castadiva running program The jar archive contains a special set of files as shown in figure 6 5 The structure of the jar file is depending on the structure of the Castadiva source folder and it contains several files e lib PluginCastadiva is related to the ServiceLoader class and must point to the class regarding packages where the plugin should be loaded In our example it would contain the following line castadiva Plugins olsrd e MANIFEST MF is an automatically generate file e olsrd class contains the plugin implementation It will be described further on e PluginCastadiva class is the ab
61. ints Get i XPos i 10 i YPos i 10 i Position cannot be outside of the canvas XPos i XPos i Float X YPos i YPos i Float Y NodeCheckPoint checkPoint new NodeCheckPoint float XPos i new Float 0 System out println Checkpoint node i j XPos YPos nodes i j checkPoint 86 float YPos il
62. ion planer s process 6 2 1 4 Validation As commented in the introduction I could not validate the original version of the Execution Planner as I did not succeed to run it Taken in account the absence of any comment in the code I would say the Execution Planner was in a beta stage It might have functioned but probably in a very precise situation with very precise arguments The execution planner has suffered many changes since the initial version and we dis covered new problems later on The difficulty with the execution planner is that it needs to take in account any possibility of Castadiva Plugins routing protocols types of traf fic All those functionalities are spread all over the original program I mean that it is not sufficient to call a loadCastadiva function and later on startSimulation There are numerous settings that have to be separately taken in account As I also commented in the introduction Alvaro made a regular use of the execution planner for his own work Most of the time it worked just fine and the benefits of the execution planner are really significant when the amount of simulations increases Never theless with over 200 runs spread over various simulations a problem occurred and I did not manage to solve it until now It might be something related with memory overflow 6 2 2 Routing Plugins 6 2 2 1 Introduction With MANETs mobility can make links among nodes unstable It is important to have stro
63. it relies on expensive clusters to emulate the sce nario Taking the limitations of these platforms into consideration Castadiva relies instead into a cheap architecture based on low cost devices to generate a test bed Castadiva outperforms the other emulators in terms of a the variety of devices that can be used as nodes b the initial budget needed to deploy the emulator and c the ease of use provided by it s GUI this means that it makes a clear contribution to the research community To our knowledge none of the emulators mentioned above support QoS traffic 3 Castadiva Castadiva is a MANET test bed platform wich provides a cost effective alternative to simu lation tools This can be achieved by allow ing certain critical components of a simulation to be real For example Castadiva relies on actual wireless communications using IEEE 802 11 interfaces OpenWrt Admin Console Core with Java application Test bed device Test bed device 1 Eth Test bed device Eth Figure 1 Castadiva architecture Castadiva s architecture is shown in Fig ure 1 It is based on two major elements a The core which orchestrates the simula tion and coordinates the Nodes b The nodes which are able to communicate among them selves for simulation purposes using their IEEE 802 11 wireless interface Communica tion among the core and the Nodes is pe
64. ite about I think it was an interesting approach that forced me to reenforce the basements of my knowledge about Castadiva I finally wrote the following sections of the article 3 Castadiva which is a presentation of the test bet 4 3 Plugin system comments the routing and mobility plugin systems When we got a first version of what the article could be we submitted our draft to Carlos T Calafate Juan Carlos Cano and Pietro Manzoni Their revision was very helpful they corrected many grammatical errors but they also pointed out some semantic errors For example we should not name Castadiva before we describe it This might of course seem obvious but it is not that simple to correctly explain something you have been working on for months It thus was an interesting exercise and I tried to remember of it for this report I also remembered LyX17 a IATEXedition tool that I discovered for the occasion and that I appreciated a lot Our article is part of the section XXI Jornadas de Paralelismo See the following URL for further details http cedi2005 ugr es 2010 contenido php apartado actividades amp sub simposios 16See figure 3 1 on page 13 for time management details See www lyx org 49 7 Useful software 7 1 Netbeans Netbeans is an IDE Integrated Development Environment which supports JAVA the language Castadiva s Core is written with It is a very efficient tool which really empowers the programming process It
65. ition float ZPosition After the Generate Plug in button was pressed the program will try to compile the code If there is any compilation error it is displayed in a special window The second part of the Mobility Plugin System takes place in the Simulation window First of all the Mobility Plugin should be available in the Mobility Model drop list Then it is important to configure the different parameters that are used in the Mobility Plugin Min Speed Max Speed Pause and Simulation time Finally pressing the New Simulation button will show the node s movement on the canvas Be aware that this is not a live replay See appendix A for more information about Castadiva s GUI 36 Mobility Model Designer void ObtainNodePositionsForEntireSimulation NodeCheckPoint nodes APs accessPoints Float minSpeed Float maxSpeed Float simulationPause int totalTime float X float Y System out printin Variables are nminSpeed minSpeed maxSpeed maxSpeed totaltime totalTime Width X Height Y 4 Array to place the nodes loat XPos new float accessPoints Size float YPos new float accessPoints Size a Initial position is 200 for all nodes or int i 0 i lt accessPoints Size i XPos i 200 YPos iJ 200 4 At each second position is incremented for al nodes or int j 0 j lt totalTime j for int i 0 i lt accessPoints Size i AP p acces
66. le Node 1 TARGET REACHED 540 0 1200 0 75 06782928526441 Km h ns_ at 52 18759071966314 node_ 1 setdest 540 0 1200 0 75 06782928526441 and converts it to a Mobility Matrix which knows for every node its position at any second of the simulation It is important to notice the granularity of the Mobility Matrix one second which is different from the granularity of the CityMob output file This will lead to approximations In order to generate the Mobility Matrix we will need to take in account that a move ment can be interrupted by another movement instruction Therefore while reading each instruction we will write it entirely to the mobility matrix If no later instruction inter feres with the current instruction the matrix is consistent If a later instruction interferes it will will consider the position of the node at the instructions time rounded to one sec ond Here we have an approximation that is transmitted along the whole simulation The importance of such an approximation can be seen in figures 6 12 and 6 13 on the facing page It is probably possible to obtain strictly correct results but it would require important changes on the original program In order to reduce the error I simply added the following Tt would be a matrix with numberOfNodes rows and totalSimulationTime columns The same matrix is used for Random Mobility and Mobility Plugins 40 One node s X position in a random simulation dependin
67. lete all Average Throguhput 0 Kb s Close Figure 4 4 Castadiva s traffic window When a simulation ends the traffic window is automatically displayed figure 4 4 That window has two purposes it can be used to configure the communications among nodes and later on to display simulation results For a more detailed overview of the Castadiva User Interface which reflects its features refer to Appendix A on page 59 Section 6 1 on page 21 can also help for comprehension it 18 explains Castadiva through a practical example The Castadiva Journal 1 is also a good source for further information 19 5 Objectives My objectives for the internship were based on a previous work realized by Ignacio Munoz Vera in 2009 Basically I had to validate his work making sure that the results generated with his new functionalities were correct and consistent But as we did not know how far his project had been carried out it was also possible that I would have to fix eventual errors The main novelties I had to validate were the following e The Execution Planner allows to setup a list for automatic simulations e The Routing Plugin System should enable the user to automatically start and stop any routing plugin on his Access Points e The Mobility Plugin System should allow to generate mobility scenarios e The CityMob Importation should allow to import a CityMob random mobility sce nario Ap
68. lity Plugin system s dynamism comes from the the possibility to use variable parameters like the amount of nodes the maximum speed in the algorithm and so on The following section will describe the design usage and validation for the Mobility Plugin System 6 2 3 2 Design As for Routing Plugins the Mobility Plugin System is based on the ServiceLoader class which allows to insert code into a running program Figure 6 8 on the next page shows how to create a jar plugin file that can be loaded using the ServiceLoader class The process is quite similar to the one described for routing Plugins 6 2 2 4 on page 29 but it is slightly more complicated regarding two points 1 The MobilityPlugin Wrapper class has some dependencies with Castadiva s files There fore we need to copy those files compile them and insert them in the jar package To avoid any confusion for Integrated Development Environment like NetBeans which scan folders for java files those files are stored as tzt files 2 The Mobility plugin system is based on a wrapper A wrapper that is nothing more that the final java file with a few code and name tags that should be replaced with the user code As for routing Plugins mobility Plugins are automatically generated and packaged They rely on the jar and javac system commands 39 C B E iS ServiceLoader User information required static Java algorithm Dependencies files Plugin name
69. mlessly deleted at any time I just created a new routing protocol plugin but it does not work prop erly There can be many reasons for a plugin not to work Nevertheless there are a few recurrent things you can try First consider the plugin as the execution of the following command Your specific commands can be found in CASTADIVA in structionsForNodeX cp castadiva nfs olsrd conf etc olsrd conf Start routing protocol usr sbin olsrd d 0 Wait for the end of simulation sleep 70 Kill the protocol killall olsrd 2 gt dev null Clean protocol configuration file rm etc olsrd conf Try to run each command separately with ssh to find out where the problem is The routing protocol is correctly activated but simulation results are not consistent Consider that most routing protocols require time to exchange routing information and establish routes Try to configure traffic a reasonable time after the simulation starts 33 6 2 2 7 Validation As previously commented the original program could not be validated as it was unable to generate any plugin A deep analysis of the code revealed several errors 1 Paths were hard coded into the program Therefore I introduced global variables to define the execution environment and give more flexibility to the program 2 Some folders where required for the plugin creation and were not automatically generated The whole structure for plugin creation was finally
70. n or simulation Further information for more detailed comprehension can be found directly into the ExecutionPlanner java file as I fully commented any piece of code I worked on 6 2 1 3 Usage There are two options to add a simulation to the execution planer s simulation list You can create a new scenario 1 or load a previously saved one 2 1 If you decide to create a new scenario the simulation window will be displayed and you will be able to setup the scenario to your needs You will then need to save it what the execution planner does basically is to use Castadiva s load function for each simulation of it s list If you do not save the scenario Castadiva wont be able to load it 2 To load a previously saved scenario you need to select either the folder containing your scenario or the scenario folder itself in the file selector When a scenario was successfully loaded it is displayed in the simulation window Be careful that if you make any changes on the scenario they wont be kept if you simply press the Accept button You need to use the save option and erase the previously loaded file or create a new one in order to keep your changes Important note In points 1 and 2 I introduced how the execution planner saves and loads scenarios Consequently you should also notice that if you use the execution planner to load any scenario your previous settings and scenario will be erased Once a scenario is loaded a new lin
71. n the execution planner ServiceLoader class usage in Castadiva Structure of a routing plugin s jar file ServiceLoader plugin generation in Castadiva 004 Routing plugin configuration window Creation of a jar ServiceLoader Mobility Plugin Mobility Plugin configuration window in Castadiva CityMob 2 0 User Interface CityMob importation process One node s X position in a random simulation without error correction One node s Y position in a random simulation without error correction One node s X position in a random simulation One node s Y position in a random simulation Low CBR UDP throughput for a mobile scenario in Castadiva ans ns 2 Distance between nodes for a mobile scenario UDP CBR Throughput comparison between ns 2 and Castadiva Test architecture for UDP routing on wireless routers Part I Personal Experience 1 Background and motivations The ERASMUS program allows two European universities to settle an agreement and to exchange their students for one or two academical semester s Achieving such an exchange is probably a unique experience in one s life It gives the opportunity to discover a new culture The student is dropped in a totally new environme
72. nWRT operating system To do so we will use a cross compiler that can be found on the OpenWRT site Make sure that the compiler corresponds to your version In the previously downloaded archive the cross compiler is located at staging_dir toolchain mipsel_gcc3 4 6 bin mipsel linux gcc mipsel linux gcc DDEBUG CASTADIVA bin UdpFlowClient c CASTADIVA bin dacme c o CASTADIVA bin UdpFlowClientMIPS mipsel linux gcc CASTADIVA bin UdpFlowServer c o CASTADIVA bin UdpFlowServerMIPS Finally the compiled files must be placed in the NFS_SERVER_PATH bin directory Note that there are MIPS and X86 files this stands for the node s processor s architecture 6 2 Validation of new functionalities 6 2 1 The execution planner 6 2 1 1 Introduction An overview of the current execution planer s Graphical Users Interface GUI can be found in appendix A on page 59 It was designed to allow automatic execution of various and possibly different simulation scenarios The user can generate a list of previously saved scenarios indicate how many times each simulation should be run and let the system process his list Results for each simulation are written in a text file The version I was asked to validate offered almost the same interface than the one shown in appendix A Unfortunately I did not manage to get it working it is to say to load a scenario nor to run the simulations All my intents were solved with some JAVA errors I thu
73. nd 109 and exceed the simulation time of 109 100 110 99 seconds With UDP as the client do not know about the reception of its packets it wont transfer more than 100 seconds During the 99 seconds out of the simulation time nothing will be transmitted Results should thus be correct but this generates a useless waiting time With TCP the client will need to connect to the server before it can transfer its data and it will start counting down its transfer time when it gets its first connection with the server As a result during the 99 seconds out of the simulation TCP will continue transferring data It is not to mention that visibility is reset after the simulation time This causes wrong results 45 Low CBR UDP Throughput comparision for a determined scenario 50 40 2 ao x 30 5 Qa lt 2 20 lt 10 0 0 2 4 6 8 10 12 Simulation progress Seconds Throughput with Castadiva Throughput with ns 2 Figure 6 16 Low CBR UDP throughput for a mobile scenario in Castadiva ans ns 2 2 With TCP If a transfer is planned for 100 seconds and if the nodes get in range for only 2 seconds at second 98 the throughput will be calculated over 2 seconds This is not taking in account problem 1 If another transfer is planned for the same 100 second and if the nodes get in range for the two first seconds the throughput will be calculated over 100 second As a result two nodes that transfer the same amount of da
74. ng routing protocols to reconfigure routes in a short time and allow efficient commu nication In Castadiva there are two options for routing 1 Static routing it is to say that routes are calculated with Castadiva for the whole simulation and automatically set on the nodes For now Castadiva only supports the OPTIMUM static routing which is based on the Dijkstra algorithm 2 Dynamic routing In such case nodes are responsible for the routing process In other words each access point must run its own and appropriate routing software Routing is thus depending on the routing software available for the Access Point s platform Point 1 is implemented since the very beginning of Castadiva It is based on some route instructions that are sent to the Access Points Point 2 was also available in legacy versions of Castadiva the user could simply install and run routing protocols on all of his Access Points Routing Plugins allow something simple but also very useful When a routing plugin is configured it receives the instructions of how to start and stop a specific routing protocol on the Access Points It can also replace one configuration file on the Access Points 28 LU L plugin jar ServiceLoader AbstractPlugin Create instance Lh 4 Use functions Figure 6 4 ServiceLoader class usage in Castadiva My work here was to validate a previously implemented Routing Plugin System Un fortunately it was absolutely not functional f
75. ni my supervisor He introduced me to the team and especially to Jorge who guided my first tasks I had to setup my work environment and get used to the software that I was suposed to validate Two weeks later on February the 10 I met Pietro Manzoni angain and we set my main objectives for the internship My work was theoretically quite simple validate a few new functionalities of a wireless simulation test bed Nevertheless we did not know exactly how far their developpement had been taken As shown in Figure 3 1 those objectives finaly spread over the whole internship with a few additionnal works like the participation to a resarch article This is due to the fact that the functions I had to validate were left in a very early developpement stage with no documentation and none of those funcionalities acutally functioned out of the box My work finally consisted in a circle of investigating fixing errors trying to validate and investigating again for the new errors Each time I focused on a new part of the software I had to analyse and understand it During my first days the program apeared to me as something huge I would never understand Slightly my knowledge of the program increased and I started assimilating pieces of the puzzle After 5 month I felt a very important progression of knowledge and effectiveness Through the present report I would like to transmit that experience 3 2 Time management Something important regarding t
76. nt with few landmarks forced to review his bounds and to open himself to new habits and a different educational system It is an opportunity that in my point of view can t be ignored For the past 3 years I have been studying a computer science degree in Toulouse The two first years of one s academical career are for obvious reasons not appropriate for an exchange During my third year I reoriented my studies from pure computer science to networks and telecommunications As I then entered a new school I probably was in a too tricky situation to consider the overload of an exchange Never the less I undertook the preparation of such a project Achieving exchange is not that common in my home school Consequently there was no specific exchange agreement settled with the university of Valencia or even with any other fully Spanish speaking university Setting up such an agreement from scratch was quite an heavy task and requested several months of administrative procedures Finally the reason why I choose Spain for my ERASMUS among other countries is that I wanted to learn a second foreign language which I believe is interesting for my professional career IThere is a partnership settled with the university of Barcelona but I was told that Catalan was the main language there 2 The Host University 2 1 Universidad Polit cnica de Valencia UPV POLITECNICA DE VALENCIA The UPV is a Spanish public university oriented to
77. o out of range nodes Linksys WRT54GL with the OpenWRT Linux based Operating system 2 Asus eee PC 901 do not communicate even if their real inter faces are able to do so The graphical user in terface also allows picking a routing protocol and mobility instructions Those possibilities will be further described in this paper Finally figure 4 is a capture of the traffic window Traffic among nodes can be config ured with a few options related to UDP or TCP such as the UDP rate or the TCP file size to transfer 4 Proposed improvements In this section we explain all the new features added to Castadiva These features include both QoS integration and automatization of testing parameters This section is divided in four parts the first two parts describe how we added QoS support to Castadiva and the other two describe the automatization and scheduling of a batch of tests In particular the first part explains how we included support for IEEE 802 11e the sec ond part explains how we managed to measure delay in UDP packets Part three describes the new plugins system that allows switching between routing protocols and mobility mod els Finally in the fourth part we describe the new execution planner and it s statistics collection system 4 1 IEEE 802 11e support IEEE 802 11e is an extension to the IEEE 802 11 standard which was released in 2005 to add QoS support QoS is achieved by re placing the DCF function
78. ome concluding remarks 2 Related work The idea of creating an emulator for MANET environments is not new and in fact several platforms for generating real ad hoc network experiments have been proposed and can be found in the literature We will only focus on those platforms that similarly to Castadiva operate with real code since our main purpose is to test and validate real code ORBIT is an indoor radio grid emulator for controlled experimentation and an out door field trial network for end user evalu ations in real world settings 4 This em ulator needs an expensive noise generator since it emulates high node distances by reducing the signal to noise ratio It also requires investing a high budget to cre ate the grid of nodes each computer is a possible position of the node in a simula tion as well as extra support servers for data storage Thus deploying the entire infrastructure requires a lot of room Carnegie Mellon U Wireless Emulator supports real devices applications MAC and PHY layers on a network wide scale while maintaining experimental control and repeatability 5 The disadvantages of this emulator are that it does not use commercial off the shelf devices using an FPGA for digital emulation instead MobiEmu an emulator to test an ad hoc net work of any scale and with any mobility scenario without actually moving the ad hoc nodes physically 10 We discarded this emulator for our test since
79. ools to use emulators or to use test beds 4 2 Introduction to Castadiva Castadiva is a MANET emulator which provides a cost effective alternative to simulation tools This can be achieved by allowing certain critical components of a simulation to be real For example Castadiva relies on real wireless communications using IEEE 802 11 interfaces The description is based on my work in a previous article 2 See Appendix B 15 Simulation L 7 v Access Point FE Lors ba ide Ye G Rate Pd Uep Core with Java application Test bed device Eth Coordination Network Test bed device Eth Test bed device Eth Figure 4 1 Castadiva s architecture Castadiva s architecture is shown in Figure 4 1 It is based on two major elements a The Core which orchestrates the simulation and coordinates the Nodes b The Nodes which are able to communicate among themselves for simulation purposes using their IEEE 802 11 wireless interface Communication among the Core and the Nodes is performed over a typical Ethernet network Ethernet allows reduced delays and guarantees no interferences with wireless signals 16 Castadiva Core Castadiva Wireless Node Figure 4 2 Software components for Castadiva As shown in figure 4 2 the Castadiva core is written with JAVA and can be run on a simple computer The Nodes on their part must rely on a Linux opera
80. or some reasons I will describe later in this section I thus reviewed the whole system in order to obtain what we just described Important note Routing Plugins do not install any routing software on the Access Points It is mandatory to previously install the adequate routing software on every Access Point The only thing routing Plugins can do is to send the start and stop instruc tions for the routing protocol and replace one configuration file on the Access Points 6 2 2 2 Routing on OpenWRT With OpenWRT based Access Points a few routing protocols are originally supplied among the original packages For example we have OLSR Babel B A T M A N BGPv4 BGPv4 BGPv4 IS IS OSPFv2 OSPF v8 RIP RIPNG It is also possible to compile some more protocols for OpenWRT Kamikaze but this is a quite difficult and random process 6 2 2 3 OLSR OLSRT Optimized Link State Routing is one of the routing protocols available for Open WRT and designed for MANETs It is also the protocol that was used for the routing plugin system s development 6 2 2 4 Design The following describes the technical implementation of Routing Protocol Plugins in Cas tadiva First of all you should know that the design of the Routing Plugin System is quite complicated compared to what it actually does This is probably because of the Mobility Plugin System See section 6 2 3 The Mobility Plugin System has higher requirements than the Routing Protocol Plug
81. ow to stop the routing protocol Castadiva will automatically find out the name of the binary file stated in A and use a killall binaryFileName instruction on each node Mobility We also extended Castadiva s to define custom mobility models during a simulation Through the graphical interface shown in figure 7 a user can take advantage of the JAVA programming language to calculate the position of each node at each second of the simulation Given sim ulation information Access Points list mini mum speed the user can write any desired code to meet the only requirement fill in the NodeCheckPoint i j array where i represents Mobility Model Designer void ObtainNodePositionsForEntireSimulation NodeCheckPoint nodes APs accessPoints Float maxSpeed Float pause int totaltime for int j 1 j lt totalTime j change the position of all nodes in one second for int i 0 i c aps Size i AP p aps Get i NodeCheckPoint checkPoint new NodeCheckPoint float p x j float p y new Float 0 nodesfi li checkPoint XTrans Generate Plug in Figure 7 Mobility plugin configuration window an access point and j a second in the simula tion Note that every access point must have a defined position at any time of the simula tion If the code is correct the plugin will be compiled and integrated directly into Cas tadiva thus becoming available in the Simu lation windo
82. provements in that version is the support of packet loss due to interferences on wireless links Another improvement is the native support of 802 119 Fortunately one of the example scripts adhoc_tcp tcl that can be found in the ns allinone 2 84 dei80211mr 1 1 4 samples folder precisely simulates an Ad Hoc 802 11g net work That script inspired my final ns 2 settings Those settings can be found in appendix F on page 80 7 2 4 Setting ranges with ns 2 In Castadiva ranges can be configured through the user interface See appendix A on page 59 If the distance among two nodes is lower or equal to their range they can communicate without any distance related loss If the distance is greater then Access Points can not communicate at all In order to set the range of a signal in ns 2 the threshold cc file located at ns indep utils propagation threshold cc can be compiled and used as follows threshold m TwoRayGround fr 2400000000 ht 1 hr 1 300 distance 300 propagation model TwoRayGround Selected parameters transmit power 1 58489 frequency 2 4e 09 transmit antenna gain 1 receive antenna gain 1 system loss 1 transmit antenna height 1 3See http www isi edu nsnam ns tutorial index html See http www isi edu nsnam ns 5See http www isi edu nsnam ns ns documentation html See http www marquez barja com See http www isi edu nsnam ns doc node223 html 51 receive antenna height
83. r formed over a typical Ethernet network Eth 63 64 Castadiva Core Castadiva Wireless Node Figure 2 Software components for Castadiva ernet allows reduced delays and guarantees no interferences with wireless signals As shown in figure 2 the Castadiva core is written with JAVA and can be run on a sim ple computer The nodes on their part must rely on a Linux operating system For exam ple Castadiva s development team used Wi Fi routers or even Netbooks Netbooks of fer better hardware performances than routers and can be easily moved around Figure 2 offers a detailed overview over the protocols used in Castadiva It represents the situation where the Core would be a Linux PC and the nodes are wireless routers using a Linux based operating system OpenWRT Since the Node devices may be very specialized routers PDAs and have limited storage resources we use NFS to store and access any simulation content on the nodes The NFS server must be located on the Core allowing easy file sharing among the Castadiva software and every node Finally Secured Shell SSH is a good option to control the simultaneous beginning of a simulation on every nodes Figure 3 shows the graphical interface that was designed to easily control the previously described system As we can see it is possible to place each node on the canvas and to de fine its range According to those ranges Cas tadiva makes sure that tw
84. re Castadiva s client and server programs that are used on the Access Points to generate TCP or UDP traffic flows See Protocol Binaries in figure 4 2 and section 6 1 3 for more information about those programs We are now going to detail how those programs calculate throughput especially how they calculate the time for the throughput calculation Please note that the following explanations might be difficult to understand They are written for someone who already 44 knows about Castadiva s implementation The two points at the end of this section shows the consequences of such an implementation with mobility With TCP the server program is started from the very beginning of the simulation The parameters for the server programs are use castadiva nfs bin TcpFlowServerMIPS lt Port gt lt secs gt lt times gt secs represents the time that the server should wait for connections before it exits For example for a simulation of 120 seconds if we have a transfer from second 10 to 60 then secs 120 The client program is started exactly when the data transfer should start It will try to connect to the server and then transmit for the determined time In our example it is started at second 10 and will transfers for 60 10 50 seconds When the server receives the first connection from the client it will record the transfer start time and continue receiving as long as he has not received for secs seconds To calculate throughput the ser
85. s started looking for what could be wrong and finally reviewed most of the source code Alvaro also participated that task as he needed the execution planner functionality for his own work Particularly he added the deferred simulation start system and the Save Load list function 6 2 1 2 Design I tried to understand and conserve as much as possible the existing code and design of the execution planner Figure 6 2 shows a very general overview of the current execution planer s functioning 5See http downloads openwrt org kamikaze 8 09 2 brcm 2 4 OpenWrt SDK brem 2 4 for Linux i686 tar bz2 25 ExecutionPlannerGUI Generate Simulations button pressed ExecutionPlanne CastadivaModel StartExecutionPlannerSimulation LoadScenario CastadivaController AllSimulationSteps CheckSimulationThreads PrintTraffic EndExecutionPlannerSimulation Figure 6 2 Global design of the Execution Planner LoadScenario and AllSimulationSteps are two main functions of Castadiva What the Execution Planner does is to take advantage of Castadiva s Save and Load function alities in order to load and process many simulations in a loop Once the scenario is loaded and when its simulation ends we need to intercept its ending instructions CheckSimula tionThreads and write the results to a file PrintTraffic Eventually we do also need to start the next ru
86. sPoints Get i XPos i 10 i YPos i 10 i jI Position cannot be outside of the canvas XPos i XPos i Float x YPos i YPos i Float Y NodeCheckPoint checkPoint new NodecheckPoint float XPos i float YPosfi new Float 0 Suctara out nrintin Chacknaint noda tit tit LB ne K HAND ne LR H F diagonalTrans Generate Plug in Figure 6 9 Mobility Plugin configuration window in Castadiva 37 and that it is not synchronized with the real simulation After a simulation was successfully processed you can review each second of the simulation using the Show feature Important note Even if the maxSpeed parameter is not set in the algorithm it is manda tory to set a positive Max Speed in the Simulation interface Otherwise mobility is not used 6 2 3 4 Validation As previously commented the original program could not be validated as it was unable to generate any plugin A deep analysis of the code revealed several errors 1 Some variable names were displayed with different case and sometimes different names in the graphical interface The variable pause was also proposed but not present in the code 2 When several plugins were available in the plugin list whatever plugin was selected the first plugin alphabetical order was always loaded and used This came from a bad usage of the ServiceLoader class I corrected this issue by cleaning up and recoding the Plugin structure
87. sciences and technology It is spread over three campuses in the region of Valencia Spain It also regroups 37000 students 2800 teachers and 2200 professionals to offer a large variety of services During the first semester of the academical year I experienced the UPV as a student and my global impression was quite positive In particular I appreciated how students were considered as the center of any concern classes did generally regroup a maximum of 20 30 persons and teachers organized tutor sessions in their office twice a week The whole forward planning of the semester was very often announced from the very beginning which allowed efficient personal organization The university also provides numerous services like a very efficient and unified intranet many sport facilities conferences libraries places to study and so on 2 2 Grupo de redes de computadores The network research group GRC was founded in 2000 and it regroups researchers from the Computer Engineering Department Currently the group s efforts are focused on Mobile Ad Hoc Networking MANETs and more precisely on the management of such networks For example it is concerned about routing security and quality of service in MANETs Figure 2 1 shows the organization of the GRC research group The faculty regroups several professors mainly working at the UPV but also coming from other Spanish uni versities Professors have several tasks like research education and administr
88. self instvar ifq_ return ifq_ paramO T Node MobileNode instproc getPhy param0 self instvar netif_ return netif_ param0 set noisePower 7e 11 set per new PER per loadPERTable80211gTrivellato per set noise noisePower set val CSThresh expr noisePower pow 10 sensingTreshdB 10 0 set val AffectThresh expr noisePower Transmission power is 32dBm Phy WirelessPhy set Pt_ 1 58489 Phy WirelessPhy set freq 2437e6 Phy WirelessPhy set L_ 1 0 The RX treshold determines the range If the received signal strength is greater than this threshold the packet can be successfully received Processed with threshold cc This threshold is for 300 m with 32dbm transmission power Phy WirelessPhy set RXThresh_ 9 90556e 10 Mac 802 11 set bSyncInterval 20e 6 Mac 802_11 set gSyncInterval_ 10e 6 Mac 802_ 11 set ShortRetryLimit 3 Mac 802_11 set LongRetryLimit_ 5 81 Mac 802 11 Multirate set RTSThreshold 100000 Mac 802 11 Multirate set dump interf_ 0 per set debug 0 PowerProfile set debug 0 Phy WirelessPhy set debug 0 Mac 802 11 Multirate set debug 0 Phy WirelessPhy set CSThresh_ val CSThresh EZZEZEZZZIZEZZZZZZZZZZZZZZZZZZZZZZZZZZZZZZ ZSZ ZSZ ZSZ SZ ZSZ ZIZI E End of the default header EEE ZSZ 82 G Simple Gnuplot example file 83 set terminal svg size 600 400 enhanced fname arial fsize 11 butt solid set output CityMob PositionX svg set key horizontal se
89. ser does not have to compile or package anything This is done au tomatically by Castadiva using the javac and jar system commands This supposes that the javac and jar are available on Castadiva s host system 6 2 2 5 Usage How to setup a routing protocol on an Access Point depends on its operating system On an OpenWRT operating system OLSR can be installed using the following commands opkg update opkg install olsrd Nevertheless there are a few points that should be considered on every system to guar antee compatibility with Castadiva e The routing protocol should not start automatically with the Access Point In our example the following command allows to enable or disable the protocol on router s startup etc init d olsrd ENABLE DISABLE Note that it is also possible to con figure startup settings in etc rc d e The routing protocol must be configured for the correct interface With OLSR on an OpenWRT operating system the interface can be selected in etc config olsrd e It might be interesting to backup the original protocol configuration file While using a plugin for the first time Castadiva may erase that file It is also interesting to use the original configuration file as a basis for custom configuration With OLSR on an OpenWRT operating system the olsrd conf configuration file is located in var etc olsrd conf To ease compilation of new Routing Protocol Plugins Castadiva offers a Graphical User Interface
90. simulation 1400 1200 1000 800 600 Distances m 400 200 0 20 40 60 80 100 120 140 Simulation progress Seconds Maximum range for a node Distance between nodes 3 5 Distance between nodes 2 4 Distance between nodes 4 5 Distance between nodes 3 4 Figure 6 17 Distance between nodes for a mobile scenario and if a node receives the command to transfer some data to another out of range node it will make an ARP_REQUEST and wont get any response After a short time it will make another ARP_REQUEST As long as the ARP_REQUEST gets no response the waiting time increases If two nodes get in range later in the simulation and if the client is still waiting for a reply to his ARP_REQUEST transfer wont start before the next ARP_REQUEST On the contrary in Castadiva it starts immediately The difference of time can lead to a difference of up to 128 seconds Important note With UDP ns 2 keeps sending ARP requests as long as it gets no an swer 6 4 Maximal UDP bit rate issue As explained in section 7 2 3 on page 51 one of my first concerns to compare Castadiva with ns 2 was to setup ns 2 so that it behaves similarly to my 802 11g network Therefore I made a speed comparison between Castadiva ans ns 2 with UDP Constant Bit Rate traffic Results can be found in figure 6 18 on the next page When I presented my first results to the group some commented the very low transfer rat
91. stract class that olsrd class implements e olsrd conf has no direct relation with the plugin system olsrd conf is the configura tion file for the routing protocol and I put it into the archive to have a unique plugin file The class file naturally comes from a java source file which must match a previously established pattern In Castadiva the pattern is an abstract class named PluginCastadiva It can be found in the lib package and suggests the following methods e getBin Where is the binary file located e getFlags What arguments should be used to run the binary file getPathConf Where is the configuration file located getConf How is the configuration file named e getKilllnstruction How to stop the protocol from running The user must implement those methods for the routing protocol he wants to use To do so the Custom Routing Protocol user interface allows the user to set String values for each of those function except for the kill instructions to stop a running protocol Castadiva uses the following command killall protocol where protocol is the binary file passed as getBin See figure 6 6 8See Javadoc for further information and tutorials 30 ae si ServiceLoader User Information lugin java E lugin class SE I lt a E gt p ugin J pag a ai required static User Interface Compilation Packaging files Figure 6 6 ServiceLoader plugin generation in Castadiva Important note The u
92. t Seeding Random number generator with val seed n ns random val seed proc usage argvO puts Usage argv0 puts tmandatory arguments puts t t x MAXX y MAXY puts toptional arguments puts t t cp conn pattern sc scenario nn nodes puts t t seed seed stop sec tr tracefile rp protocol n proc getopt argc argv global val lappend optlist nn nn traffic sources sc rp trfile httpfile stopns pagesize objsize lista de parametros de ejecucion de sim ns parametro valor otroparametro valor 71 100 101 102 103 104 105 107 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 127 128 129 130 131 132 133 134 135 136 137 for set i 0 i c argc incr i set arg lindex argv i if string range arg 6 0 continue set name string range arg 1 end set val name lindex argv expr i 1 If routing protocol is DSR adjust queue type if val rp DSR amp amp val ifq Queue DropTail PriQueue set val ifq CMUPriQueue to set one session per traffic source set val nn_session val nn_traffic_sources puts n n n puts gt Configuration puts gt set val chan Channel WirelessChannel channel type puts gt set val prop Propagation TwoRayGround radio propagation model puts gt set val netif Phy WirelessPhy network interface type
93. t title One node s X position in a random simulation depending on simulation progress set xlabel Simulation progress Seconds set ylabel Node s X coordinates set yrange 0 1500 plot positionNode tr using 1 2 title NS node 1 with lines positionNode0 txt using 1 2 title Castadiva node 1 with lines positionNodel tr using 1 2 title NS node 2 with lines positionNodel txt using 1 2 title Castadiva node 2 with lines positionNode2 tr using 1 2 title NS node 3 with lines positionNode2 txt using 1 2 title Castadiva node 3 with lines positionNode3 tr using 1 2 title NS node 4 with lines positionNode3 txt using 1 2 title Castadiva node 4 with lines positionNode4 tr using 1 2 title NS node 5 with lines positionNode4 txt using 1 2 title Castadiva node 5 with lines unset output 84 H Mobility Plugin algorithm for the Mobility Plugin System System out println Variables are nminSpeed minSpeed maxSpeed maxSpeed totaltime totalTime Width X Height Y Array to place the nodes float XPos new float accessPoints Size float YPos new float accessPoints Size Initial position is 200 for all nodes for int i 0 i lt accessPoints Size i XPos i 200 YPos i 200 At each second position is incremented for al nodes for int j 0 j lt totalTime j for int i 0 i lt accessPoints Size i AP p accessPo
94. ta over the same time get very different results The nodes that first enter in contact are disadvantaged for TCP throughput calculation I will now present my last validation results that lead me to the present conclusions In figure6 16 we have a comparison of a low CBR Constant Bit Rate scenario in Castadiva and ns 2 It is the same scenario imported from CityMob and commented in figures 6 14 and 6 15 on page 43 in the CityMob Import validation see section6 2 4 4 In that scenario the following transfers were configured 1 2 1 3 1 4 1 5 2 3 2 4 2 5 3 4 3 5 4 5 from second 10 to 110 In figure 6 16 they are numbered from 1 to 10 in the same order The first observation that we can make is that visibility is correct When ns 2 can transfer Castadiva can also transfer Figure 6 17 shows the distance between the different nodes still for the same simulation For readability only nodes that get in range a least once in the simulation are shown This confirms correct mobility management in Castadiva retarding visibility as the higher throughput are associated with the nodes that spend more time in range 6 3 2 Bad ARP emulation with TCP Castadiva s visibility rules do not block ARP_REQUESTs Therefore two theoretically out of range nodes do constantly know their respective MAC addresses even tough they can not communicate at layer 3 In ns 2 if there is no special routing protocol enabled 46 Distance between nodes for a mobile
95. tcortes batousay com jorgehortelano gmail com calafate jucano pmanzoni disca upv es Abstract Nowadays Mobile Ad Hoc Network MANET testing is being done using simulation models like ns 2 but such simulations tend to be too optimistic In this work we extend Castadiva a test bed architecture that allows validating software solutions both applications and net work protocols for real ad hoc environments using low cost off the shelf devices and open source software Our extensions are twofold in the one hand we improve the platform to enhance support for real time traffic On the other hand we make distributed tests possible and fully automatic 1 Introduction Mobile ad hoc networks MANETs are packet radio networks composed by independent and heterogeneous stations that cooperate in rout ing and packet forwarding tasks conforming a dynamic multi hop network All the nodes which are part of a MANET can act as end points for data interchange or as routers when end points are not in direct contact The topology of the network changes dynamically as mobile nodes join or depart the network or when radio links between nodes become unusable and so networks can be de ployed easily and cheaply by using efficient routing protocols to operate correctly The importance of MANETs becomes more evi dent by noticing the wide application area that MANET scenarios embrace Special situations require communication networks to be ava
96. ted in the ns 2 s code beneath the comment see code in section 6 2 4 2 As ns 2 handles meters seconds there probably is something wrong For now we have configured Castadiva so that it behaves similarly to ns 2 6 2 4 3 Usage There is no particular requirement to import a CityMob scenario to Castadiva As it is the current version we used CityMob 2 for the development of this functionality Importation might no longer work if the CityMob output file structure undergoes any change Once you generated the scenario use the import gt CityMob menu and select CityMob s output file If you want to set custom parameters for the imported Access Points configure the configuration aps tzt file in Castadiva s folder to do so Each line of that file represents an Access Point the CityMob importation system will consider line one for it s first Access Point and so on Finally note that a CityMob mobility model is not saved when the save function is used in Castadiva You will thus need to load the CityMob file before you load a scenario 6 2 4 4 Validation The original CityMob import did not work out of the box It assumed hard coded that the nodes where named with a string followed by a number from 0 to N 1 For example my nodes where named from 1 to N as recommended in the Castadiva Journal 1 This was corrected simply removing this requirement and calling nodes with their number instead of their name Castadiva s results were
97. th to the NFS folder was not loaded when a scenario was loaded As a result if a non default path was used it had to be reconfigured manually after each load 3 The Net device value in the Computer configuration window was not updated after a new scenario was loaded 4 The Stop simulation button was no longer compatible with mobility I thus replaced it with a Reset Access Points which should send a reboot command to all the Access Points 5 The buttons Import Citymob and Import NS 2 in the execution planner had no associated code They were removed 6 2 5 2 Unfixed issues 1 The Net device value in the Computer configuration window not seem to be used in the program 2 The export NS 2 function exports TCP traffic as UDP TCP is not implemented in the code 6 3 Important mobility issues in Castadiva While trying to compare Castadiva s results with ns 2 s results for data transfers with mobility none of my comparisons were conclusive During my last days investigating the possible origin of the problem I found an error with mobility in Castadiva 6 3 1 I also found an issue with ARP that can be problematic in certain situations 6 3 2 Unfortunately I had no sufficient time to fix those problems and this should probably be done as soon as possible as the problem also probably affects the current release of Castadiva 6 3 1 Bad throughput calculation Originally Castadiva was designed for static scenarios And so we
98. then compared to ns 2 for a random UDP scenario results were very different Indeed the original importation misunderstood CityMob it assumed that when a movement instruction was given at time T to move from position A to position B in C seconds the node would necessarily be in position B at time T C In fact a node can receive new instruction during its travel In such case it interrupts its current movement and starts moving according to the new instructions Correcting this made me review a very large portion of the code Investigating the existing code was a hard task as there were no comments I thus fully commented the CityMob import system In order to validate my results I finally compared Castadiva s new node positions to ns 2 s ones As shown in figures Figure 6 14 on page 43 and Figure 6 15 on page 43 results are now quite conclusive Note that the slight differences come from Castadiva s sampling which only occurs every second Important note The CityMob importation is functional and correct but the results of any subsequent simulation can not be considered as valid This is because of a problem that was discovered in the Access Point s client and server programs in particular with TCP This problem is affecting mobility in Castadiva since it s very beginning and is detailed in section 6 2 3 4 on page 38 13 See http www isi edu nsnam ns tutorial section IX 2 42 Node s X coordinates Node s Y coordinat
99. ting system For example Castadiva s development team used Wi Fi routers or even Net books 7 Net books offer better hardware performances than routers and can be easily moved around Figure 4 2 offers a detailed overview over the protocols used in Castadiva It represents the situation where the Core would be a Linux PC and where the nodes would be wireless routers using a Linux based operating system OpenWRT Since the Node devices may be very specialized routers PDAs and have limited storage resources we use NFS to store and access any simulation content on the nodes The NFS server must be located on the Core allowing easy file sharing among the Castadiva software and every Node Finally Secured SHell SSH is a good option to control the simultaneous start of a simulation on every Nodes 2LinkSys WRT54GL with the OpenWRT Linux based Operating system 3 Asus eee PC 901 17 4 3 Castadiva s features iz CASTADIVA SIMULATION JEE Si mulation om 200 m 400 m 600 m 800 m 100 m 1200 m 1400 n p Access Point Bounds router2 x x 1500 0 m Paa ted aopen 00 Position v 1500 0 m x 712 0 m Time 4 amp Simulation 5 of arouter3 Total a al DS 600 m WiFi Connection arouterd Sa Range 400 0 m Node Mobili de Mob y oe le O RTS CTS Z Show Max speed 0 0 0 0 mys Pause oo s E Mobility Model a ee View M Grid RAN
100. traffic is exported as UDP See Unfixed issue 2 on page 44 Nevertheless the export function is very interesting when the simulation has many nodes with complicated movements 7 3 Gnuplot Gnuplot is a powerful tool that allows to generate any kind of graphical charts For my validation work I often had to compare the results of a Castadiva simulation with the results of a ns 2 simulation Some of those charts can be viewed in this report for example Figures 6 12 and 6 14 Those figures were exported directly in vector graphical format SVG which offers many advantages as an excellent image quality a light weight and infinite scalability As for ns 2 Gnuplot changes a lot along versions and it is not always easy to find the adequate information But the Gnuplot user manual has an index which makes it quite 3 See http evanjones ca basic 80211 stats html The following legend of ns 2 output format might be helpful http nsnam isi edu nsnam index php NS 2_Trace_Formats 10 See www gnuplot info See http www gnuplot info documentation html 52 easy to use As an example of how to generate a chart with Gnuplot you can refer to appendix G on page 83 which is the source file of figure 6 14 7 4 ETEX I discovered ATEX during this internship I first used it to write the Routing Plugin Systems user manual ATEX has the advantage of producing high quality documents as it respects many typographical rules which are
101. ver will use the time between the first and the last received packets With UDP the server program is also started from the very beginning of the simulation The parameters for the server program are use castadiva nfs bin UdpFlowServerMIPS lt Port gt lt waiting secs gt lt receiving seconds gt lt times gt Where waiting secs indicates how long the server should wait for connections and receiving seconds shows how long the reception should last The client programs starts simultaneously with the transfer The client program transfers for the desired time and then exits As it is UDP it wont try to get any confirmation from the server When the server receives its first packet it starts counting for receiving seconds before it shuts down Now there is a problem when both those server programs are used with mobility As it is possible that two nodes get in range and out of range at any time two problems can arise 1 If nodes are out of range when the transfer starts In a static simulation they remain out of range for the whole simulation With a mobile simulation they might get in range later in the simulation If they do so Servers do only start counting transfer time from the moment the first packet is received For example if a simulation lasts 120 seconds and if there is a transfer among two nodes from second 10 to second 110 If the nodes are out of range until second 109 they will start transferring for 100 seconds at seco
102. w s mobility plugin drop list see Figure 3 4 4 Execution planner The first version of Castadiva had had the fol lowing drawbakcs tests should be performed one by one causing the creation of a huge set of experiments to be very inefficient With this new feature we have managed to run up to 200 tests in a row allowing us to collect and process lots of statistics The execution planner alone is very power ful but if we combine it with the new plugin system Section 4 3 we can automatize the ex periment launching process varying not only the traffic sources destinations but also the routing protocol or the mobility model Thus we consider that it is a good platform for eval uating ns 2 results and performing compar isons As we can see in figure 8 the GUI of the ex ecution planner provides an easy way to cre ate and automatize the MANET evaluation It includes features like Load list which im ports a list of scenarios previously saved with the Save list option The main component of the window is a table that reflects the dif 67 68 CASTADIVA Execution Planner Source folder Results folder Runs Status Ihome alvaro ScenTODOJEF Ihome alvaro ScenTODO EF _ 9 Cancelled jhome alvaro ScenTODO EG lhome alvaro ScenTODO EG 10 Ready Jhome alvaro ScenTODO EH Jhome alvaro ScenTODO FG home alvaro ScenTODO FH home alvaro ScenTODO EH 10 Ready Jhome alv
103. w set f3 open out3 tr w set f4 open out4 tr w set f5 open out5 tr w set f6 open out6 tr w set f7 open out7 tr w set f8 open out8 tr w set f9 open out9 tr w Create the general operations director Used internally by MAC layer must create create god val nn LL set mindelay_ 50us LL set delay 25us LL set bandwidth _ 0 not used LL set off prune 0 not used LL set off CtrMcast_ 0 not used Agent Null set sport_ 0 Agent Null set dport_ 0 Agent CBR set sport_ 0 Agent CBR set dport_ 0 Agent TCPSink set sport_ 0 Agent TCPSink set dport_ 0 Agent TCP set sport_ 0 Agent TCP set dport_ 0 Agent TCP set packetSize 1460 unity gain omni directional antennas set up the antennas to be centered in the node and 1 5 meters above it Antenna OmniAntenna set X_ 0 Antenna OmniAntenna set Y_ 0 Antenna OmniAntenna set Z 1 5 Antenna OmniAntenna set Gt_ 1 0 Antenna OmniAntenna set Gr_ 1 0 Initialize the SharedMedia interface with parameters to make 78 it work like the 914MHz Lucent WaveLAN DSSS radio interface Phy WirelessPhy set CPThresh_ 10 0 Phy WirelessPhy set CSThresh_ 1 559e 11 Phy WirelessPhy set Rb 11 1e6 Phy WirelessPhy set freq _ 914e 6 Phy WirelessPhy set L_ 1 0 Phy WirelessPhy set bandwidth_ 11le6 Phy WirelessPhy set bandwidth 54e6 Initialize the 802 11 MAC Configuration for 802 11a and 802 11g Mac 802 11 Mac 802 11 Mac 802 11 Mac 802 11 Mac 802 11 Mac 802 11 Mac 802

Download Pdf Manuals

Related Search

Related Contents

Manual PT-0083 BICICLETA BLUE LINE  Samsung ES20 Manual de Usuario  Operation Guide  Forte 3 User Manual  Laser Tools Copyright Laser T ht Laser Tools Copyright Laser Tools  SONY α NEX-5, NEX-3  JVC KS-FX250 Cassette Player User Manual  HDR110 - Restroom Direct  Manuel d`installation rapide et de mise en route  

Copyright © All rights reserved. Failed to retrieve file