as a PDF - Informatik IV
Contents
1. 30 Property window of MAP_EM eee eee 31 Property window of MAP_MOEA 32 Property window of JMomfit 33 Property window of ACfit 34 Property window of MEP IT 000000 0G 34 Property window of ARIMA Fitting 35 Property window of ARTAT IG s cit y a Yeh ee ed 36 Property window Of CAPP P it era o SSS ee we a 36 Property window of Modgen a e ee o 37 Property window of Plot 2 2 ni eee ee 38 Property window of Print 0 00 5002 ee 39 Property window of Queue aaa a a a 40 1 Introduction ProFiDo Processes Fitting Toolkit Dortmund is a flexible environment written in Java for realising different steps of input modeling ProFiDo offers a GUI to combine var 10us command line tools realising steps of data preprocessing parameter fitting and analysing the resulting processes into a workflow In the current version effective August 2014 ProFiDo supports fitting of common distributions Phase type PH dis tributions Markovian Arrival Processes MAPs Correlated Acyclic Phase Type Pro cesses CAPPs AutoRegressive To Anything ARTA processes and AutoRegressive Integrated Moving Average ARIMA models from trace data and the analysis of these distributions and processes The theoretical background on these stochastic processes is out of scope of this manual Readers not familiar with these p2. Figure 10 Positioning the output node 11 such that the node is highlighted Afterwards right click on the destination node and select the first option Connect output from selected node to this see Fig 11 which results in a directed arc connecting source and desti nation node see Fig 12 In a similar way all other arcs can be created completing the workflow specifi cation see Fig 13 Of course arcs may be created at any time during the workflow specification not only after all the nodes have been placed Each arc contains a label with the model type for the model description that 1s propagated between the two nodes Additionally a file name is associated with each connection ProFiDo will automatically determine an unique file name for each arc such that no conflicts occur However in some cases it is desirable to manually chose a file name For example the tools Plot and Print use the file names as labels for the curves of the plot or for the columns of the table Observe from Fig 13 that we changed the file names at the arcs from JMomfit and ACfit to Plot and Print to MAP3_JM xml and MAP3_AC xml The file name can be changed by selecting Edit Connection from the context menu of the arc 600 ProFiDo File Workflow Edit Settings Help Moments 5 JMomfit Connect output from selected node to this Delete Edit Parameters Edit Comment Moments Change Name tes State Iterations Show Command
3. ibt3_dat trace O Figure 4 Positioning the input node 3 In a similar manner other nodes can be placed on the canvas E g to insert the call of a fitting tool into the workflow click on Workflow in the menu bar and select the specific type of job using Add Job see Fig 5 After positioning the node a window will pop up where job specific parameters can be specified see Fig 6 For the example workflow two MAP fitting algorithms are added to the work flow in particular a tool called ACfit that fits a MAP according to the lag k autocorrelations of the trace and a tool called JMomf it that uses the empirical joint moments for fitting Since both tools expect a PH distribution as input we place the tool Momfit that fits a PH distribution according to the moments of a trace on the canvas Since we want to assess the quality of the fitting algorithms we add the tools Plot and Print to the workflow Plot can create figures showing differ ent properties of the models like autocorrelations or the distribution function Print has a similar function for statistical properties that should be output in textual format and can for example generate a IATEX table Fig 7 shows the canvas with all tools added so far Note that the tool specific parameters that are entered in the property window from Fig 6 are displayed above the different job nodes Thus Momfit will use the first 9 empirical moments from the trace to fit a PH distri
4. 100 vi Enable vi Visible v Format lt seed gt Result 13 13 vi Enable vi Visible OK Last Values Defaults Cancel Figure 29 Property window of Momfit e Input Trace e Output PH distribution MAP e Parameters Moments The number of empirical moments of the trace that should be considered for fitting 30 States The order i e the number of transient states of the PH distri bution The algorithms stops 1f either the number of Tterations has been reached the difference between the results of two subsequent iterations is smaller than Epsilon or Time has been reached Seed Seed of the random number generator that 1s used to determine the initial random PH distribution 4 4 2 Fitting of Stochastic Processes ProFiDo currently supports two types of stochastic processes namely MAPs and ARI MA ARTA CAPP models and a variety of different fitting algorithms for these pro cesses MAP_EM MAP_EM 8 applies an EM approach to the whole trace The approach is time consuming so that it should not be applied to fit MAPs with a larger state space to long traces However since EM algorithms have a monotonic convergence the algorithm can be used to improve the likelihood of a MAP that has been fitted by some other more efficient approach In case only a trace file is provided as input the algorithm will start with a random MAP if additionally an existing MAP is provided as input the algo
5. 4 2 2 Building a Workflow seis Sock es wt ee SW eS Be cae we 18 4 2 3 Executing a Workflow 2 2004 21 AS ProFiDo s Meny Str ct re nilo os Bee Se a ee ee 23 ASA TheFile Menu rri rraian EA Oo Eee a 23 4 3 2 The Workflow Menu aaa a 24 Tao dhe Cdt Men ss ita E a de E E 25 4 3 4 The Settings Menu a 25 439 Ame Help VERLA An de a ess 26 AO Context Menus ss wa di paisa BE doe ee ae 26 44 Supported TOONS 2 e as aon Se eee Book wh Ok ee es E A e 27 4 4 1 Distribution Fitting o 27 4 4 2 Fitting of Stochastic Processes 24 4 31 44 3 Trace Generauon 4 24 o a A a eS 37 44 4 Statistical Two Sample Tests 37 4 4 5 Result Visualisation 0 38 A Developing ProFiDo 41 A l Integrating additional tools 41 Ad JODINOGUES gt 032060 Ad EA A E 41 A 1 2 Input and Output Nodes 41 das Test Mana er 4 4 ia A A A ee SE ee 42 A 2 1 TestManager class 2 4 24i2 4044842644584 4445 42 A 2 2 Bash scripts perform_tests sh and perform_diffs sh 43 References 45 Index 47 List of Figures O 0 10D UE WN _ ae SS lt gt BR BW WWW WW WWW WNN NNN NNN WN WN e FR PR RRR Re OOO JOYA UD oOoOowo0oo J 0O8SMN PWNFR TO WONADW MN E Y WN Empty WORIOW esas a ic ey Ses do are od 6 Adding an input file node T Selecting
6. GUI Windows Right click profidol 3_win zip and select Extract from the files con text menu and choose a target folder the recommended directory is C Note that the identifier of the selected target folder must not contain a blank or other white space characters Ihttp www mingw org wiki msys Afterwards the selected directory will hold a subfolder called MinGW which contains the preconfigured MinGW MSY S environment and the entire ProFiDo software Open the subfolder MinGW msys 1 0 within the target directory and execute msys bat to starta MSYS shell Within this MSYS shell type cd profido to change to the ProFiDo directory and run profido sh to start ProFiDo s GUI 3 Quick Start Guide The following sections explain step by step how ProFiDo s GUI can be used to build up an example workflow description and how this workflow can be executed This quick start guide will only explain the basic functionality that 1s necessary to specify the example workflow and gives a first overview of the features of ProFiDo A detailed description of ProFiDo that introduces all the features and properties will be given in Sect 4 3 1 Specification of an Example Workflow In the following we will assume that the user wants to create a simple workflow to fit MAPs to a given trace file using two different fitting approaches To assess the quality of the fitted MAPs some characteristics of the MAPs and the trace file should be
7. ADP Edit Job G FIT Single Erlang Format lt m gt lt r_1 gt lt r_2 gt lt r_m gt Result 3 1 1 1 3111 _ Enable Visible All Sia Format lt N gt lt m_min gt lt m_max gt Result 5 1 5 515 v Enable vi Visible Format lt eps gt Result 0 00005 0 00005 M Enable Visible Conv Check T Set vi Visible Logarithmic Trace aggregation Format lt Ita gt Result 100 100 _ Enable Visible Uniform Trace aggregation Format lt lua gt Result 100 100 Enable Visible OK Last Values Defaults Cancel Figure 28 Property window of G FIT e Input Trace e Output PH distribution MAP e Parameters Single Erlang Fits a single hyper Erlang distribution to the trace The parameter is specified as a list of numbers where the first number de termines the number of Erlang branches and the following numbers define the number of phases for each branch e g 3 1 1 1 fits a hyper Erlang distribution with 3 branches where each branch consists of 1 phase All Erlang Fits different settings of hyper Erlang distributions with a given number of states and selects the distribution with the best likeli hood value The parameter is specified as a list of 3 values where the first value determines the overall number of states the second value the min imal number of Erlang branches and the third value the maximal number of branches Note that only
8. Figure 20 Property window of the Edge View Plot 4 2 3 Executing a Workflow If the workflow description is complete it can be exported to a bash script by selecting Export Workflow as Script from the Workflow menu Before the script is 21 600 Setup edge view Print Format Format lt Format gt Result format txt txt Y Enable Relative Format lt relative gt Result false false Y Enable Start Format lt start gt Result 0 0 Y Enable End Format lt end gt Result 10 10 v Enable Type Format lt Type gt Result plot cdf cdf gt Y Enable OK Last Values Defaults Cancel Figure 21 Property window of the Edge View Print generated ProFiDo checks the workflow for consistency A correct workflow has to fulfill the following conditions A workflow has to be acyclic A workflow must start with at least one input node and terminate with at least one output node Each input node must have at least one outgoing edge Each output node mut have exactly one incoming edge Each job node must have at least one incoming and one outgoing edge If these conditions are met a file dialog pops up where the user has to enter a file name After entering a filename ProFiDo will generate two shell scripts lt name gt sh and lt name gt _setup sh The execution of lt name gt _setup sh will prepare the workspace 1 e setting up all the tools used in the workflow After that lt
9. directly With the help of the Plot Details and Print 20 Details menu items 1t 1s possible to display intermediate results 600 ProFiDo File Workflow Edit Settings Help Format ps Start 0 Al Erang 515 End 10 Conv Check Set Type pdf Plot Delete Connection Edit Connection Add EdgePoint Delete EdgePoint v Auto Position Show XML File Plot Details Print Details Figure 19 Plotter functionality It should be noted that the script cf Sect 4 2 3 must have been exported and executed once before to use the Plot DetailsandPrint Details functional ity A detailed description of the tools Plot and Print will be given in Sect 4 4 Clicking on the menu item Plot Details a window similar to the property win dow of the job node P Lot will pop up see Fig 20 The behaviour of the item Print Details is analogous see Fig 21 With a click on the OK button the Plot Print 1s executed with the edge file as input and the characteristics of the selected model are displayed in a new window To ensure that everything works as described the workflow must be saved after exporting Otherwise you will be prompted to specify the path of the exported script 600 Setup edge view Plot Start Format lt start gt Result 0 0 v Enable End Format lt end gt Result 10 10 v Enable Type Format lt Type gt Result plot cdf cdf Y Enable Ok LastValues Defaults Cancel
10. ficients that are considered for fitting min AR order max AR order min MA order max MA order Determines the order of the base process CAPP Fit will fit base processes for all possible combinations given by the minimal and maximal AR and MA order and select the base process with the best result 4 4 3 Trace Generation Modgen This job node allows for the generation of a trace from a given model de scription Only a single model description is allowed as input The trace length can be defined by setting parameter TraceLength Fig 38 shows the parameters of the node Modgen e00 Edit Job Modgen TraceLength Format lt observations gt Result 10000 10000 Ml Enable M Visible OK Last Values Defaults Cancel Figure 38 Property window of Modgen e Input Distribution PH distribution MAP CAPP ARIMA or ARTA process e Output Trace e Parameters TraceLength determines the length of the generated trace 4 4 4 Statistical Two Sample Tests Dist_Tests Job node Dist_Tests applies statistical two sample tests offered by the statistical software R 23 in this way validating the hypothesis that two samples are from the same distribution The result is a IEX table giving the P values of the applied tests At present Kolmogorov Smirnov and Pearson s Chi Squared tests are supported Job node Dist_Tests needs no parameters specified by the user and exactly two inputs 1 e two input arcs of type XML Trace are
11. Processes INFORMS J on Computing 10 1 72 81 1998 R Hornig and A Varga An Overview of the OMNeT Simulation Environ ment In Proceedings of 1st International Conference on Simulation Tools and Techniques for Communications Networks and Systems SIMUTools 2008 45 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 W D Kelton and A Law Simulation Modeling and Analysis McGraw Hill 2000 Jan Kriege and Peter Buchholz An Empirical Comparison of MAP Fitting Al gorithms In Bruno Mu ller Clostermann Klaus Echtle and Erwin P Rathgeb editors Proceedings of the 15th International GI ITG Conference on Measure ment Modelling and Evaluation of Computing Systems and Dependability and Fault Tolerance MMB amp DFT 2010 volume 5987 of LNCS pages 259 273 Springer 2010 Jan Kriege and Peter Buchholz Correlated phase type distributed random num bers as input models for simulations Performance Evaluation 68 11 1247 1260 2011 Marcel F Neuts A versatile Markovian point process Journal of Applied Prob ability 16 764 779 1979 Marcel F Neuts Matrix geometric solutions in stochastic models Johns Hopkins University Press 1981 Structured Markov Matrix Market http www4 cs tu dortmund de buchholz struct matrix market html OMNeT Discrete Event Simulation System Version 4 0 User Manual Andriy Panchenko Modelling of Network Processes
12. are ac for lag k auto correlation coefficients cdf for the cumulative distribution function pdf for the probability density function mom for moments and jmom for joint moments Queue The Queue node is an additional way to compare processes This job node simulates a single server queue with a trace or stochastic process as input and returns a plot of the queue length distribution The unit of measurement for the interarrival times used as input for the queue are in seconds Fig 41 shows the tool specific prop erties The Queue node expects the service time the buffer size and the simulation time as additional parameters The job node uses OMNeT 12 as simulation engine and therefore the service time must be specified in OMNeT s syntax see below e Input any number of Traces MAPs CAPPs ARTA and ARIMA processes e Output Image in PostScript The value given for queue length i is the sum of all time intervals where customers have been in the queue divided by the overall simulation time Note that the given queue length distribution might vary for different simulation times because of transient behaviour Note that no confidence intervals are calculated For further information on simulation see e g 13 39 ANDO Edit Job Queue Service Time Format lt service time gt Result exponential 1 0s exponential 1 0s v Enable vi Visible Buffer Format lt buffer gt Result 10 10 vi Enable v Vi
13. background For the interested reader some references providing this background information will be given 4 4 1 Distribution Fitting ProFiDo currently supports three tools for distribution fitting DistFIT is a general tool for fitting various common distributions whereas GFit and Momfit specialise on fitting PH distributions 27 DistFit DistFit provides parameter estimation for various common distributions based on a given trace For most distribution types the fitting is based on the MLE maximum likelihood estimator 13 A complete list of supported distributions their properties and the corresponding fitting approach can be found on the ProFiDo website 5 Fig 27 shows the parameters of Dist Fit Depending on the parameters entered 600 Edit Job DistFit Distribution Type Format lt Distribution Type gt Result auto auto Y Enable Y Visible OK Last Values Defaults Cancel Figure 27 Property window of DistFit the tool fits a single distribution or several distributions to the trace and determines the one with the best likelihood value The output of the tool is an XML file as specified in the ProFiDo XML interchange format 6 It contains the distribution with the best likelihood value for the given trace as its first distribution tag All other fitted distributions if any are written as sub tags of a first level lt info gt tag sorted in order of their likelihood value Furthermore each lt distribut
14. coefficients of a trace this approach constructs an ARMA base process such that the resulting Correlated Acyclic Phase Type Process CAPP with marginal distribution Fy exhibits the desired autocorrelations from the trace For more details see 15 Fig 37 shows AOO Edit Job ARTAfit Autocorrelations Format lt ac gt Result 10 10 v Enable m Visible min AR order Format lt pmin gt Result 2 2 M Enable Visible max AR order Format lt pmax gt Result 5 5 mM Enable m Visible min MA order Format lt qmin gt Result 0 0 mM Enable vi Visible max MA order Format lt qmax gt Result 2 2 m Enable v Visible OK Last Values Defaults Cancel Figure 36 Property window of ARTAfit the parameters of the node CAPP Fit e00 Edit Job CAPP Fit Autocorrelations Format lt ac gt Result 30 30 v Enable vi Visible min AR order Format lt pmin gt Result 3 3 m Enable v Visible max AR order Format lt pmax gt Result 7 7 m Enable v Visible min MA order Format lt qmin gt Result 2 2 m Enable v Visible max MA order Format lt qmax gt Result 4 4 m Enable v Visible OK Last Values Defaults Cancel Figure 37 Property window of CAPP Fit e Input PH Distribution Trace e Output CAPP e Parameters 36 Autocorrelations Determines the number of autocorrelation coef
15. description of this job Edges e Delete Connection Deletes the connection e Edit Connection Opens the attribute window of the edge where the file name file type and order key can be changed see Sect 4 2 for an explanation of these attributes Additionally one can specify which of the attributes should be displayed on the canvas e Add EdgePoint Adds an additional point to the edge that can be used to specify the pathway of the edge e Delete EdgePoint Deletes the edge point provided it is not the sole point of that edge e Auto Position If enabled the selected edge point will be positioned auto matically otherwise it can be positioned manually within the grid e Plot Details Opens the edge view plot window that can be used to plot characteristics of the model that is propagated along the edge e Print Details Opens the edge view print window that can be used to print characteristics of the model that is propagated along the edge 4 4 Supported Tools ProFiDo currently supports several tools for fitting and analysing distributions and stochastic processes that can be integrated into a workflow as job nodes cf Figs 5 and 23 In the following these tools and their parameters will be introduced Ad ditionally the generated output and the expected inputs are listed to clarify which of the job nodes may be connected We will only briefly introduce the basic ideas of the different algorithms but omit the full theoretical
16. file types and vice versa By encapsulating the actual command line tools in small hand written bash scripts even tools that don t allow the specification of file names by command line parameters can be integrated in ProFiDo An example of this approach can be seen in the integration of G FIT cf 1 A 1 2 Input and Output Nodes As job nodes and their connections are used to represent the execution and data prop agation of command line tools an explicit representation of external files that are en tered into and taken from the workflow is required External files are represented by input and output nodes By defining converters for input and output nodes ProFiDo is able to manage the conversion from the external file types to the common exchange format and vice versa 41 A 2 Test Manager In order to detect potential errors a test suite allowing automated tests of the ProFiDo GUI has been developed Its central approach is based on comparing exported work flow scripts of a modified ProFiDo version with corresponding reference exports thus detecting major errors and inconsistencies caused by any changes made It consists of two main parts which are described in the following A 2 1 TestManager class The TestManager superclass within ProFiDo s source code provides a standardised ar chitecture which allows the simulation of any desired user interaction In combination with ProFiDo s ability to load a workflow from the locatio
17. for all models and the values relative to the moments of the first model are plotted For jmom the joint moments E X X7 are used Print The Print node offers the possibility to generate IATEX tables containing several characteristic figures of process descriptions Currently supported are moments and joint moments Fig 40 shows the tool specific properties e Input any number of Traces distributions PH distributions MAPs CAPPs ARIMA and ARTA processes e Output Table in I4IFX format cf Table 1 38 ADP Edit Job Print Format Format lt type gt Result latex ry latex v Enable Visible Relative Format lt relative gt Result false false M Enable Visible Start Format lt start gt Result 0 0 vi Enable vi Visible End Format lt end gt Result 10 10 W Enable M Visible Type Format lt Type gt Result plot jmom jmom B v Enable v Visible OK Last Values Defaults Cancel Figure 40 Property window of Print e Parameters Format File format of the generated output Currently only latex is supported Relative If true the Joint moments are computed relative to the values of the first model 1 e the model resulting from the job node that has been connected first to the Print node If false the absolute values are printed Start and End determine the first and last value to be computed Type Type of data to be printed Possible values
18. name gt sh can be called to execute the actual workflow lt name gt sh contains all the calls to the tools corresponding to the job nodes of the workflow ProFiDo will automatically determine the right order of the calls from the workflow layout After the execution 1s complete the output files can be found in the location specified by the output nodes of the workflow Interim results can be found in directory Results The file names in this directory conform to the parameters of the node connections In addition to the normal export an extended workflow with additional report can be created After selecting Export Workflow as Script with Report ProFiDo will export the above described workflow scripts and two additional files lt name gt _report sh and lt name gt _report tex The lt name gt _report sh script will be called af ter the execution of the Workflow and create a versatile PDF report file containing 22 all properties of the workflow and its intermediate as well as final results All files referenced and used by the report are in directory Results Report Note that re port files lt name gt _report sh and lt name gt _report tex are directly generated from ProFiDo s GUI and are not generated by setup scripts Note that the Results and the Report directories are not automatically deleted by the generated scripts It is in the user s responsibility to save files that are still needed from the folders and to delete them b
19. one of the parameters Single Erlang or All Erlang may be enabled 29 Conv e Determines the e value for the convergence check If the values between two iterations of the EM algorithm improve less than e the optimisation stops Conv Check Determines the type of convergence check If checked the maximal difference between parameter values of the distribution is used otherwise the difference between log likelihood values Logarithmic Trace aggregation If enabled this value specifies the number of intervals that are used to aggregate the trace on a logarithmic scale Uniform Trace aggregation If enabled this value specifies the number of intervals that are used to aggregate the trace Momfit Momfit 9 fits the parameters of an acyclic PH distribution by applying a non linear optimisation algorithm according to the moments of the trace The optimi sation approach minimises the least squares difference between the weighted moments of the trace and the fitted distribution and is independent of the length of the trace Fig 29 shows the property window of Momfit ANA Edit Job Momfit Moments Format lt moments gt Result 5 5 v Enable vi Visible States Format lt states gt Result 3 3 vi Enable vi Visible Iterations Format lt iter gt Result 10000 10000 M Enable M Visible Epsilon Format lt eps gt Result 1 0e 7 f 1 0e 7 v Enable v Visible Time Format lt time gt Result 100
20. the random initial MAP Autocorrelation determines the number of lag k autocorrelations that are used for MAP fitting JMomfit 9 performs a least squares fitting of the weighted joint mo ments of the trace and the fitted MAP It is started with a PH distribution that is ex panded into a MAP Consequently the node expects a trace and a PH distribution as input Since only linear least squares problems have to be solved and joint moments are fitted the approach is very efficient and the effort 1s independent of the length of the trace The parameter window is shown in Fig 32 600 Edit Job JMomfit Joint Moments Format lt jmoments gt Result 3 3 v Enable vi Visible OK Last Values Defaults Cancel Figure 32 Property window of JMomfit e Input Trace PH distribution e Output MAP e Parameters Joint Moments The number n of joint moments to be considered for fitting The algorithm will use the first n x n joint moments from the trace ACfit ACfit 14 takes a trace and a PH distribution as input and fits a MAP ac cording to the lag k autocorrelations of the trace This implies the use of a non linear optimisation method so that the effort becomes slightly higher than that for joint mo ment fitting but the effort is still independent of the trace length Fig 33 shows the parameters of ACfit 33 ANA Edit Job ACfit Autocorrelation Format lt acorr gt Result 10 10 v Enable vi
21. 3 bash script 21 24 CAPP 3 36 CAPPfit 36 Change Workflow Name 24 Chi Squared test 37 Clear History 25 Connect output from selected node to this 19 Connection of nodes 12 19 Parameter 20 Debug 26 Default file names 20 Delete History 25 directed arcs 19 Directory Set Working 24 Display Edge Information 20 26 Grid 26 Node Information 26 Dist_ Tests 37 DistFIT 28 DistFit 27 auto 28 Erlang 28 exponential 28 gamma 28 Johnson 28 lognormal 28 normal 28 triangular 28 uniform 28 distribution fitting 27 Edit Menu 23 25 enabled parameter 19 Erlang 28 Evolutionary Algorithm 32 exponential 28 Export Workflow as Script 12 21 24 Export Workflow as Script with Report 13 22 24 File Input 24 Menu 23 Output 18 24 Trace 18 file format documentation configuration 3 stochastic processes 3 workflows 3 23 File Menu 23 fitting of distributions 27 Fitting Tools ACfit 33 ARIMA 35 ARTAfit 35 CAPPfit 36 DistFIT 28 G FIT 29 insertion of 8 JMomfit 33 MAP_EM 31 MAP_MOEA 32 MEPfit 34 47 Momfit 30 G FIT 29 gamma 28 enuplot 4 GPL 3 Grid 19 26 Help Menu 23 26 History Clear 25 Delete 25 Redo 25 Show 25 Undo 25 Hyper Erlang distribution 29 info tag 28 Input file 6 Input node 18 19 Internet Traffic Archive 6 jmom 38 39 JMomfit 8 33 Job Add 24 Node 18 19 Johnson 28 joint moments 33 38 Kolmogo
22. FiDo can be used for fitting stochastic processes In Sect 4 2 the workflow specification and execution is presented Sect 4 3 gives a detailed overview on ProFiDo s menu structure and the functions available from the menu bars Finally in Sect 4 4 we give some background on the tools that are currently integrated into ProFiDo and introduce their parameters 4 1 Commandline Parameters ProFiDo can be called with various commandline parameters to load an existing work flow or to perform automated tests Most of the commandline parameters are used to control the integrated Test Manager which is explained in Appendix A in detail If no parameters are given ProFiDo will start with an empty canvas ProFiDo can be called by profido sh OPTIONS INPUT FILE EXPORT FILE where all parameters are optional INPUT FILE specifies a workflow to be loaded EXPORT FILE a file name to store the workflow OPTIONS may be any of the following e c lt file gt Selects a custom configuration file Default software xml e n Disables all file existence checks when loading a workflow e nv Disables the XML validation e h help Displays usage information 17 e t lt mode gt lt interactions gt Activates the TestManager in the selected mode and performs the specified amount of user interactions Default No mode selected interactions 10 e d lt delay gt Specifies a custom time in milliseconds between two user inter a
23. Format lt maxgen gt Result 300 300 v Enable vi Visible Alpha Format lt alpha gt Result 60 60 v Enable m Visible Mu Format lt mu gt Result 30 30 mM Enable v Visible Lambda Format lt lambda gt Result 30 30 m Enable v Visible Epsilon Format lt epsilon gt Result 1e 3 le 3 mM Enable v Visible lteration Format lt iter gt Result 100 100 m Enable vi Visible Dimension Format lt dim gt Result 5 5 m Enable v Visible Alpha2 Format lt alpha2 gt Result 5 0 5 0 m Enable v Visible Seed Format lt seed gt Result 111 111 M Enable Visible Autocorrelation Format lt acov gt Result 10 10 M Enable Visible OK Jl Last Values Defaults Cancel Figure 31 Property window of MAP_MOEA lutionary algorithm should use 32 JMomfit Alpha Mu and Lambda are parameters for the evolutionary algorithm They determine the population size a the number of parent individuals ju and the number of offspring individuals A Epsilon Stopping criterion for the EM part of the algorithm If the improvement of the algorithm is less than e the EM algorithm will stop Iteration determines the number of iterations of the EM algorithm in the first fitting step Dimension determines the order of the MAP 1 e the number of states Alpha2 is a factor used for the randomisation technique Seed of the random number generator used to generate
24. MA Fitting AR Order Format lt p gt Result 3 3 v Enable vi Visible Degree of Differencing gt Format lt d gt Result 0 v Enable vi Visible 0 MA Order Format lt q gt Result 2 2 v Enable vi Visible OK Last Values Defaults Cancel Figure 35 Property window of ARIMA Fitting e Input Trace e Output ARIMA process e Parameters AR Order Degree of DifferencingandMA Order specify the order of the ARIMA process 1 e the number of autoregressive coefficients p the degree of differencing d and the number of moving average coeffi cients q ARTAfit For a given distribution Fy and empirical autocorrelation coefficients of a trace this approach constructs an ARMA base process such that the resulting ARTA process with marginal distribution Fy exhibits the desired autocorrelations from the trace For more details see 11 Fig 36 shows the parameters of the node ARTAfit e Input Distribution Trace e Output ARTA process e Parameters Autocorrelations Determines the number of autocorrelation coef ficients that are considered for fitting min AR order max AR order min MA order max MA order Determines the order of the base process ARTAfit will fit base processes for all possible combinations given by the minimal and maximal AR and MA order and select the base process with the best result 35 CAPP Fit Fora given PH distribution Fy and empirical autocorrelation
25. ProFiDo The Processes Fitting Toolkit Dortmund Manual Falko Bause Souffian El Baba Philipp Gerloff Alparslan Kirman Jan Kriege Moussa Oumarou Informatik IV TU Dortmund profido1s4 cs uni dortmund de August 4 2014 Edit Job MAP_EM Y Enable Visible Format vV Visi ProFiDo batch 100 Y Enable Y Visible File Workflow Edit Settings Help Format lt dim gt Result 6 6 Y Enable Y Visible Al Erlang 5 15 Conv e 0 000001 Format 2 0 Y Enable Visible Format lt seed gt Result 123 123 Enable Visible OK Last Values Defaults Cancel resutt pdf XML Trace XML PDF Result pdf 1 Seite probability density function flteration 100 Map Dims 6 Alpha 2 0 Contents 1 Introduction 3 2 Installation 3 De ACENTO 1 a E OS o a a aa ee S 3 2 2 System Requirements e 3 22E Ee AAA NS TN 4 222 NAMGOWS re A UTA e ode oe a ee ce ae i 4 Zo Anstallins PIORMIDO 2 24 ed be we A Beh i bie eee 4 Doel EDU e be ete ha ee ey Oe od a ee a i 4 22 WIndows amp eead he A eR Lee Game 4 3 Quick Start Guide 6 3 1 Specification of an Example Workflow 6 3 2 Executing the Workflow 2 0 0 0 02 ee eee 12 4 Working with ProFiDo 17 4 1 Commandline Parameters 17 4 2 Workflow Specification and Execution 18 AD General Quine s s x oo amp bee ae are wk Se ee eS 18
26. Report Save Screenshot Figure 23 Workflow Menu e Add Job offers the possibility to add a new job node into a workflow For se lection one of the available fitting tools supported by ProFiDo have to be chosen The tools currently supported by ProFiDo are presented in detail in Sect 4 4 e Add Output File This item allows to insert an output node to the work flow e Add Input File This item allows for the insertion of an input node e Change Workflow Name allows to change the name of the current work flow e Set Working Directory The working folder directory can be set by clicking on this menu entry The working folder is that folder directory which will be suggested by ProFiDo for storing output files or from where to import input files e Export Workflow as Script This item allows to export a bash script which governs the execution of the specified workflow e Export Workflow as Script with Report This item exports the standard workflow script and all files needed to create the automatic report 24 e Save Screenshot This item allows to save a screenshot of the current canvas content in Portable Network Graphics PNG format 4 3 3 The Edit Menu ano ProFiDo File Workflow Edit Settings Help Undo Redo Show History Clear History Figure 24 Edit Menu The Edit Menu cf Fig 24 offers basic undo redo functionality in particular e Undo the last performed action e Redo the last cance
27. Type ac ACfit Figure 7 Canvas with several job nodes in the menu bar see Fig 8 and choose the directory where the output file shall be stored and specify its name Afterwards a window pops up where you can select the type of output see Fig 9 Finally the node representing the output file is positioned on the canvas see Fig 10 For our example we place two output nodes on the canvas one for a PostScript file 1b1_ac ps and one for a BIEX file Lb1_jmom tex eoo ProFiDo File WMO Edit Settings Help Add Job Add Output File Add Input File Change Workflow Name Set Working Directory Export Workflow as Script mfit Export Workflow as Script with Report Save Screenshot T T T Figure 8 Adding an output file 5 To complete the workflow description nodes have to be connected by edges In order to connect two nodes by a direct arc first click on the source node 10 Select Output Type s Please select type of external output file XML Trace gt Trace Cancel OK 0 0 0 ProFiDo File Workflow Edit Settings Help Figure 9 Selecting the type of output File Workflow Edit Settings Help Joint Moments 5 JMomfit Type jmom Format latex Start 1 End 5 Moments 9 States 3 Iterations 20000 Autocorrelation 10 Format ps Start 1 End 10 Type ac ACfit
28. Visible OK Last Values Defaults Cancel Figure 33 Property window of ACfit e Input Trace PH distribution e Output MAP e Parameters Autocorrelation The number of empirical autocorrelations from the trace that should be considered for fitting MEPfit From a set of moments and joint moments the algorithm constructs two matrices Ho and H as described in 24 27 and tries to transform them into a valid MAP description Dg D1 by searching for a non singular matrix B such that Do B HoB and D B H B It should be noted that the last step might fail if there is no MAP that exhibits the desired joint moments or if the method does not find such a MAP In this case the best matrices that have been found are returned which either describe a Matrix Exponential Process MEP or may not even describe a stochastic process at all Fig 34 shows the parameters of MEP fit OO Edit Job MEPfit Order Format lt order gt Result 3 3 v Enable vi Visible OK Last Values Defaults Cancel Figure 34 Property window of MEP fit e Input Trace e Output MAP e Parameters Order The number of states of the MAP 34 ARIMA Fitting The free software R 23 contains a large number of statistical meth ods including methods to determine the parameters of ARIMA models that are used in ProFiDo to generate ARIMA processes from trace data Fig 35 shows the parameters of the node ARIMA Fitting 600 Edit Job ARI
29. ails 2 2 System Requirements ProFiDo is available for Linux and Windows Depending on the operating system different requirements have to be fulfilled to run ProFiDo 2 2 1 Linux To run ProFiDo s GUI only Java is required However for full functionality including all the supported tools for fitting and result visualisation the following system require ments have to be fulfilled 2 2 2 Linux Java wish Tcl Tk shell required for converter scripts gnuplot required for creating plots of autocorrelation cdf and pdf R required for ARIMA fitting and statistical tests 23 Windows The Windows version of ProFiDo is bundled with a MSYS environment that contains most of the tools required by ProFiDo For the Windows version only Java is needed 2 3 2 3 1 P 2 3 2 Installing ProFiDo Linux Extract profidol 3 zip toa local directory the recommended directory is ProFiDo MKATE PLOriLDS nz protidol 332499 d ProriDG The directory should now contain four files o gt rofido jar software xml profido sh gpl 2 0 txt and a subdirectory software If you did not use ProFiDo as installation directory you have to edit the file software xml and replace HOME ProFiDo in line lt softwaresource gt s HOME ProFiDo software lt softwaresource gt at the beginning of the file with your installation directory Change to your installation directory and run profido sh to start ProFiDo s
30. b is defined by a set of param eters and the number of parameters might vary dependent on the job type The meaning of the parameters is also explained in Sect 4 4 The user can edit pa rameters by changing the default values Each parameter has two check boxes where the user can select whether the parameter will be enabled and or be vis ible on the grid Only enabled parameters will later be relevant when calling the corresponding tool A summary with all parameters whose visibility flag is set to true 1s displayed above the job node on the canvas Additionally all pa rameters are displayed in a tooltip when the mouse pointer is over the job node Furthermore an optional description of a parameter is displayed when the user clicks on the associated question mark cf Fig 28 e Output nodes are the counterpart of input nodes and mark the end of a work flow Thus defining output nodes is similar to the definition of input nodes Select Workflow gt Add Output File and specify name and location within the shown file selection window Afterwards you will be asked to select the output type of the file After that the output node can be placed on the grid Output nodes are used to specify the results of a workflow that the user is inter ested in For example they might be used to save the fitted models or some plots with characteristics of the models Connecting nodes To complete a workflow description the nodes of the workflow have to be co
31. bution with 3 states and run for 20000 iterations JMomfit will consider the first 5 x 5 joint moments and ACfit the first 10 lags of autocorrelation Plot will create a PostScript PS image plotting the first 10 lags of autocorrelation and Print will create a IATEX table with the first 5 joint moments A detailed description of the different tools and their parameters will be given in Sect 4 4 To define the outputs of the workflow one can specify output nodes Defining an output file is similar to specifying an input file Select Add Output File File UCSI DistFit Add Output File G FIT Add Input File Momfit Change Workflow Name MAP_EM Set Working Directory MAP_MOEA Momfit Export Workflow as Script ik Export Workflow as Script with Report MEPfit Save Screenshot ARIMA Fitting ARTAfit CAPP Fit Modgen Plot Print Queue Dist_Tests Figure 5 Inserting a job node Joint Moments Format lt jmoments gt Result 5 ur y COI File Workflow Edit Settings Help ES Enable ES visible 4 Joint Moments 3 JMomfit Figure 6 Example of a parameter window of a job node eno ProFiDo untitled File Workflow Edit Settings Help Joint Moments 5 JMomfit Format latex Start 1 End 5 Type jmom Moments 9 Print States 3 1b3_dat trace Iterations 20000 Format ps Start 1 End 10 Autocorrelation 10
32. by Means of Markovian Arrival Processes PhD thesis Fakult t Informatik TU Dresden 2007 Andriy Panchenko and Axel Thiimmler Efficient Phase type Fitting with Aggre gated Traffic Traces Perform Eval 64 7 8 629 645 2007 Vern Paxson and Sally Floyd Wide area traffic The failure of Poisson modeling IEEE ACM Transactions on Networking 3 226 244 1995 R Development Core Team R A Language and Environment for Statistical Com puting R Foundation for Statistical Computing Vienna Austria 2008 ISBN 3 90005 1 07 0 http www R project org M Telek and G Horvath A Minimal Representation of Markov Arrival Pro cesses and a Moments Matching Method Performance Evaluation 64 9 12 1153 1168 2007 The Internet Traffic Archive http ita ee Ibl gov Axel Thiimmler Peter Buchholz and Miklos Telek A novel approach for phase type fitting with the EM algorithm IEEE Trans Dep Sec Comput 3 3 245 258 2006 A van de Liefvoort The Moment Problem for Continuous Distributions Tech nical Report WP CM 1990 02 University of Missouri Kansas City 1990 46 Index lt name gt sh 12 22 lt name gt _report sh 13 22 lt name gt _report tex 13 22 lt name gt _setup sh 12 22 question mark 19 ACfit 8 33 Add Input File 6 19 24 Add Job 8 19 24 Add Output File 8 19 24 ARIMA 3 35 ARTA 3 ARTA process 35 ARTAfit 35 auto 28 Auto Position points 20 Autocorrelation fitting 3
33. command line tool and exchange format between them and does not impose any limitation on the software controlled by it The tools that can be managed with ProFiDo are specified in the software xml file The general approach of adding new job and input nodes is described below For detailed information about the configuration format please refer to the provided specification 2 A 1 1 Job Nodes The main building block of any ProFiDo workflow are the job nodes where each job node represents the execution of one command line tool All tools controlled by the GUI are required to accept at least one input file and to create at least one output file whose file names have to be definable by a command line parameter By connecting job nodes the desired data propagation of results between the jobs 1s determined and all file name management will be handled by ProFiDo When defining new jobs in the configuration file the file types accepted and cre ated by the job and the corresponding parameter have to be specified This allows the GUI to ensure that only matching job nodes can be connected In order to accomplish a more homogeneous environment it is recommended to specify an XML based ex change format between different job types as we have done in 6 By defining an input and output converter for each type of job ProFiDo is able to manage the conversion from data propagated within the workflow to the command line tools original accepted and created
34. ctions within the TestManager Default 5 e s lt seed gt Specifies a custom seed for the TestManager Default Time since midnight January 1 1970 UTC in milliseconds 4 2 Workflow Specification and Execution In ProFiDo the fitting of stochastic processes is described by workflows Workflows are composed of different nodes that represent different steps of data preprocessing fitting and result visualisation Each node of the workflow corresponds to a tool that is responsible for performing one of these steps The nodes are connected by arcs to determine the order of execution of the tools and the result propagation from one tool to the next In this way it is possible to specify complex workflows consisting of various connected tools in a graphical manner 4 2 1 General Outline A workflow 1s represented by a graph composed of a minimum of the following nodes e At least one input node associated with an input file which usually is a trace file or a stochastic process in ProFiDo s XML interchange format 6 The first entry of a trace file 1s the number of subsequent elements in that file All other entries are interpreted as data values e g inter arrival times packet S1ZeS etc e One or several job nodes each representing a particular tool Detailed descrip tions of the tools will be given in Sect 4 4 e One or several output nodes associated with an output file Possible outputs are traces stochastic processes in differ
35. e of the above set tings 1 e some of the information on edges or nodes are always displayed 1f the corresponding item from the menu is selected 4 3 5 The Help Menu The Help Menu cf Fig 26 just gives more information about the tool ANP ProFiDo File_ Workflow_ Edit Settings Help About ProFiDo Figure 26 Help Menu 4 3 6 Context Menus Some of the functionality of nodes and edges is available via different context menus that are selected by a right click on the node or on a green point of an edge Input Nodes and Output Nodes e Connect output from selected node to this This entry 1s used to connect two nodes e Delete Deletes the node e Change External File Name Selects another file for the node e Change External File Type Changes the type of the file Edit Comment Edit the comment containing a description of this file Job Nodes e Connect output from selected node to this This entry is used to connect two nodes e Delete Deletes the node 26 e Edit Opens the attribute window of the job node The attributes displayed in those windows depend on the type of job node see Sect 4 4 for all job nodes and their corresponding attributes e Change Name Changes the name of the job node that is displayed in the canvas e Show Commandline Call Prints the commandline call of the job node that 1s used when exporting the workflow to a shell script e Edit Comment Edit the comment containing a
36. ed x_1 sh x_m sh where m is the number of desired test rounds In order to detect potential errors the exported scripts are then compared to refer ence exports of the corresponding workflow For each workflow located at the specified workflow folder a reference export with the same base name is expected to be located at the specified reference location The actual comparison is done by the script perform_diffs sh which out puts an overview containing a list of all exports that did not match the corresponding reference file In order to filter differences that will only result in a different order of execution line wise alphabetical sorted versions of the input files are compared additionally Furthermore lines only containing irrelevant information e g infor mation on the ProFiDo version user specified folders etc are filtered using grep before any other processing is done For a more detailed documentation of script perform_diffs sh see the first lines of the script file Be aware that the provided scripts can be a substantial help in detecting errors but should never be relied on exclusively as only certain types of errors can be detected this way The perform_tests sh script takes a series of parameters to specify the files and folders on which it should operate and the desired amount of tests The script is called with the following options perform_tests sh lt profido jar gt lt workflow folder gt lt reference folder
37. eed of RNG 31 33 Settings Debug 26 Display Edge Information 20 26 Display Node Information 26 Settings Menu 23 25 Show History 25 tag info 28 Tcl Tk 4 Test Manager 42 tooltip at points 20 Trace file 18 Trace generation 37 49 triangular 28 two sample tests 37 Kolmogorov Smirnov test 37 Pearson s Chi Squared test 37 Undo 25 uniform 28 visible parameter 19 wish 4 Workflow 8 connecting nodes 12 description 6 edges 10 execution 12 13 17 21 23 export to shell script 12 layout 20 load 23 new 23 nodes 8 save 23 setup 13 23 specification 17 18 Workflow Menu 23 24 Working Directory Set 24
38. ent formats like ProFiDo s XML in terchange format or Nsolve s 18 format for MAP descriptions and image or TEX files for various plots and tables with characteristics of stochastic pro cesses and traces An example of a workflow is shown in Fig 13 4 2 2 Building a Workflow Specification of a workflow can be done by placing nodes on the grid and connecting them with directed arcs The connection has to observe certain rules 18 Placing nodes on the grid ProFiDo distinguishes three types of nodes Input nodes job nodes and output nodes e Input nodes are the first nodes of a workflow and have no predecessor Input nodes are used to specify files that are used as input for the workflow In most cases the input nodes will be used to load a trace file To add an input node select Add Input File from the Workflow menu Afterwards a file selection window will pop up where you can select the input file In the next dialog window the type of the file has to be chosen After that one can place the input node on the grid by clicking on the corresponding position in the canvas e Job nodes are the basic building blocks of a workflow that perform the actual fit ting tasks Each job node corresponds to one of the tools presented in Sect 4 4 To place a job node on the grid select one of the tools from Workflow gt Add Job After positioning the job node by clicking on the grid a window with tool specific parameters is shown Each jo
39. etween the execution of different work flows to avoid confusion with old existing files 4 3 ProFiDo s Menu Structure ProFiDo s GUI offers five menu entries These are e File Menu e Workflow Menu Edit Menu e Settings Menu and e Help Menu and will be described in the following Additional functionality is available via various context menus described at the end of this section 4 3 1 The File Menu The File Menu offers basic functionality for loading and saving workflows The menu is shown in Fig 22 ANA ProFiDo File Workflow Edit Settings Help New Workflow Load Workflow Save Workflow Quit Figure 22 File Menu e New Workflow offers the possibility to create a new workflow e Load Workflow allows to load a workflow description from a file in XML format as specified in 3 e Save Workflow allows to save a workflow in a file using the XML Format as specified in 3 e Quit quits ProFiDo 23 All File Menu options include a check for unsaved changes to the current work flow A confirmation dialog will pop up remembering to save the current workflow 1f necessary 4 3 2 The Workflow Menu This menu contains all the ingredients for specifying a workflow see Fig 23 It offers the following functions e090 ProFiDo Edit Settings Help Add Job amp Add Output File Add Input File Change Workflow Name Set Working Directory Export Workflow as Script Export Workflow as Script with
40. file inputfile and 20 user interactions removals of nodes and edges will be performed with a delay of 10ms between each action After all user interactions have been performed the current workflow will be exported as workflow script using the passed output file output file as target location Whenever the TestManager is activated a brief overview of the selected settings is outputted on the command line Furthermore the command line call that will execute the exact same test again is printed out gt If x is the number of user interactions min x number of nodes node deletions and afterwards min x number of edges edge deletions are performed 42 A 2 2 Bash scripts perform_tests sh and perform_diffs sh Using the above described capabilities of ProFiDo the script perform_tests sh performs a series of tests on an executable jar version of ProFiDo When executing the script all workflows located at a specified location are loaded and several test rounds are performed with every workflow each consisting of loading randomly modify ing and exporting the workflow using the workflowPanel mode of the TestManager described above The exported workflow scripts are named according to the loaded workflow and are continuously numbered during the test rounds allowing easy attri bution of input workflows to the corresponding export and test round For a workflow located at the specified location and named x sh the exported workflow scripts will be nam
41. gt lt target folder gt lt test count gt lt actions per test gt where e lt profido jar gt specifies the location of the JAR File which 1s used for the tests e lt workflow folder gt defines the location where the workflows used for the tests are located Any XML file located in this folder will be used for the test e lt reference folder gt specifies the location where the reference exports of the workflows are located As ProFiDo creates a setup script for each exported workflow script additional files named x_l_setup sh x_m_setup sh will be created 5 Additionally to the reference workflow script a corresponding reference setup script is expected as well for each workflow at the specified reference location 43 A reference export for each workflow in the lt workflow folder gt is expected to be located here e lt target folder gt defines the folder where the exported workflows and report files will be stored e lt test count gt defines the number of test rounds that should be done on each workflow file e lt actions per test gt specifies the amount of user interactions amount of deleted nodes and edges as described above that should be done within one test round The perform_tests sh script creates four detailed report files for each work flow found in the workflow folder listing all differences between the workflows ref erence exports and the exports created during the test rounds Additiona
42. ints are automatically positioned by ProFiDo To position points freely on the canvas the option Auto Position has to be disabled The parameters of a connection are the file type and file name associated with the output of a node and the order key The file type 1s displayed if the option Display Edge Information is enabled in the Settings Menu All parameters are also displayed as a tooltip if the cursor hovers over the corresponding point The order key can be used to establish the order of incoming arcs on a node If a node can process several incoming file parameters with the same type a lower order key is used first when the shell script for workflow execution is built Clicking on an existing point allows for editing the three parameters of a connection between two nodes which are the file name and the file type that are used for storing the output of a node when executing the workflow and the order key By default ProFiDo determines file names such that no conflicts will occur and the user does not have to care about the file names However some job nodes currently Plot and Print use the file names as labels in their output figures and tables See Fig 13 in Sect 3 for an example ADP ProFiDo untitled File Workflow Edit Settings Help Joint Moments 5 XML MAP JMomfit Figure 18 Complete workflow after positioning points An additional menu item see Fig 19 is offered to display characteristics of mod els eg CDF ACF
43. ion gt tag has an lt info gt sub tag containing the likelihood value of the corresponding distribution for the given trace Note that the information from the lt info gt tag is usually not used by other job nodes e Input Trace e Output Distribution e Parameters auto Fit all supported distributions to the trace and select the one with the best log likelihood value exponential Fits an exponential distribution to the trace normal Fits a normal distribution to the trace lognormal Fits a lognormal distribution to the trace johnson Fits a Johnson distribution to the trace uniform Fits an uniform distribution to the trace triangular Fits a triangular distribution to the trace erlang Fits an Erlang distribution to the trace gamma Fits a gamma distribution to the trace 28 G FIT G FIT 26 implements an EM Algorithm to fit the parameters of hyper Erlang distributions maximising the likelihood value Hyper Erlang distributions are a subclass of PH distributions with a special structure resulting in an efficient algorithm that can be applied to large traces G F IT can operate in two modes Either a single hyper Erlang distribution with a given structure 1s fitted or the best distribution with a given overall number of states is determined The parameters of G F IT are shown in Fig 28 G F IT can either use the complete trace for fitting or work on an aggregated trace as described in 21
44. line Call 1b13_ dat trace Momfit Figure 11 Connecting with destination node 3 2 Executing the Workflow Once the workflow specification is complete the entire workflow can be exported to a Shell script and be executed Select Export Workflow as Script from the Workflow menu and choose a file name to export the workflow ProFiDo will create two shell scripts lt name gt _setup sh and lt name gt sh that setup all the necessary tools and execute the tools of the workflow ProFiDo automatically sorts the order of execution of the different tools corresponding to the job nodes according to their order 12 eno ProFiDo untitled File Workflow Edit Settings Help Joint Moments 5 JMomfit Moments 9 Print 0 St 3 ates Iterations 20000 datttace ae Momfit Autocorrelation 10 Figure 12 Result after connecting two nodes in the workflow Call lt name gt _setup sh to setup the tools for the workflow After that the workflow can be executed by calling lt name gt sh In addition to the export of the normal workflow an extended workflow with ad ditional report can be created Select Export Workflow as Script with Report and choose a file name accordingly ProFiDo will export the above described workflow scripts and two additional files lt name gt _report shand lt name gt _report tex located in directory Results Report The lt name gt _report sh script will be called after the execution of the
45. lled action e show all previously performed actions via the Show History item and e delete the complete history Show History displays a list of all previously performed actions with the last action being on bottom of the list By selecting one of the previous actions one can undo all later actions by selecting Move to selected state It is also possible to redo the changes by selecting a later action afterwards The button Create Snapshot saves the currently visible state of the workflow to an XML file in the ProFiDo direc tory 4 3 4 The Settings Menu This menu cf Fig 25 allows for the definition of preferences which influence the GUI s look and feel AOO ProFiDo File Workflow Edit Settings Help Scale on Resize v Display Grid v Display Edge Information Y Display Node Information Debug e Always Display Order Keys Display All Edge Information Display All Node Information T P Figure 25 Settings Menu e The Scale on resize checkbox determines whether the graph of the work flow will be resized when the window is resized or whether it will maintain its original size 23 e Display Grid determines whether a grid will be shown or not e Display Edge Information determines whether any information are dis played on the central point of the edge e Display Node Information determines whether the visible parameters of a job node are displayed e Via the entries of the Debug menu one can override som
46. lly sorted versions of the files are compared to each other as described above Furthermore a logfile is created within the target folder listing all command line calls executed by the script in order to allow easy recreation of results when needed Assuming that a folder within the user s home directory named ProFiDo test with two subfolders exists one containing input workflows and one containing the corre sponding reference exports a sample call performing 4 test rounds on each of the workflows with 8 user interactions per test round would be of the following form perform_tests sh ProFiDo profido jar ProFiDo test workflows A ProFiDo test reference_exports ProFiDo test 4 8 44 References 1 2 3 4 5 6 7 8 9 10 11 12 Falko Bause Peter Buchholz and Jan Kriege ProFiDo The Processes Fitting Toolkit Dortmund In Proc of the 7th International Conference on Quantitative Evaluation of SysTems QEST 2010 pages 87 96 2010 Falko Bause Philipp Gerloff Alparslan Kirman Jan Kriege and Daniel Scholtyssek ProFiDo XML Configuration Format Specification 2012 http www4 cs uni dortmund de profido Falko Bause Philipp Gerloff Alparslan Kirman Jan Kriege and Daniel Scholtyssek ProFiDo XML Workflow Format Specification 2012 http www4 cs uni dortmund de profido Falko Bause Philipp Gerloff and Jan Kriege ProFiDo A To
47. n specified by its first com mand line parameter and export it to the location specified by its second parameter a series of tests can be automated by a bash script which is described in Sect A 1 Launching ProFiDo with the command line parameter t lt test mode gt lt user interac tions gt enables the TestManager and specifies the desired number of user interactions By providing the additional command line parameter d lt n gt a temporal delay between the user interactions can be specified Furthermore the software configuration file which is software xml by default can be specified with the c parameter Currently a subclass testing the main GUI by performing a series of random modi fications deleting nodes and edges to the workflow and undoing these changes after wards is provided It is selected by passing workflowPanel as test mode The nodes and edges that will be deleted are selected randomly By default the random number generator 1s using the time since midnight January 1 1970 UTC as its seed In order to reproduce errors that were encountered during previous tests it is possible to specify the seed manually by using the s parameter E g the call Java jar profido jar inputfile outputfile Y c tmp mysoftware xml A t workflowPanel 20 Sd O Ss SAS IZ 65 will start ProFiDo using tmp mysoftware xml as configuration file and seed ing the random number generator with 314159265 Upon starting the GUI loads the input
48. needed for proper functioning e Input Two Traces e Output Table in IEX format e Parameters NONE 37 4 4 5 Result Visualisation Plot The Plot node helps to visualise model characteristics such as cumulative dis tribution functions and autocorrelation lags for given model descriptions or traces Its properties are shown in Fig 39 no Edit Job Plot Format Format lt Format gt Result format ps ps B v Enable Visible Start Format lt start gt Result 0 0 vi Enable vi Visible End Format lt end gt Result 10 10 v Enable vi Visible Typ Format lt Type gt Result plot pdf vi Enable vi Visible OK Last Values Defaults Cancel Figure 39 Property window of Plot e Input any number of Traces distributions PH distributions MAPs CAPPs ARIMA and ARTA processes e Output Image in PostScript PDF or PNG format e Parameters Format File format of the generated output image Allowed values are ps for PostScript pdf for Portable Document Format or png for Portable Network Graphics Start and End determine the range on the x axis that is plotted cf Fig 14 Type Type of plot Possible values are ac for lag k autocorrelation co efficients cdf for the cumulative distribution function pdf for the prob ability density function mom for moments and jmom for joint moments For mom the moments E X for all i in the range from Start to End are determined
49. nnected by directed arcs These arcs determine the order of execution of the different jobs and the propagation of intermediate results e g if two nodes are connected the output of the first node will be used as input for the second node The connection of two nodes by a directed arc is done as follows First select the source node then right click on the target node A pop up menu will appear which allows different operations Select Connect output from selected node to this to connect the two nodes Note that two nodes can only be connected if the output of the first node is an allowed input for the second node In particular input nodes have no predecessor and output nodes no successor Input nodes may be connected with job nodes if the job node can process the input file and job nodes 19 may always be connected with an output node Connections between job nodes are allowed 1f the output of the first job matches an expected input of the second job Sect 4 4 contains all the expected inputs and outputs of job nodes Usually an arc 1s a straight line from one node to another In some cases this might not result in the best layout for the workflow e g if the arc crosses other arcs or nodes Therefore each arc comprises one or more green points which can be freely positioned on the canvas to give a more readable graph layout see Fig 18 Additional points may be added by right clicking on one of the existing points By default all po
50. olkit for Fitting In put Models In Bruno Miiller Clostermann Klaus Echtle and Erwin P Rathgeb editors Proceedings of the 15th International GI ITG Conference on Measure ment Modelling and Evaluation of Computing Systems and Dependability and Fault Tolerance MMB amp DFT 2010 volume 5987 of LNCS pages 311 314 Springer 2010 Falko Bause Philipp Gerloff Jan Kriege and Daniel Scholtyssek Distribution Overview 2014 http www4 cs uni dortmund de profido Falko Bause and Jan Kriege ProFiDo XML Interchange Format Specification 2012 http www4 cs uni dortmund de profido G E P Box and G M Jenkins Time Series Analysis forecasting and control Holden Day 1970 Peter Buchholz An EM algorithm for MAP fitting from real traffic data In Peter Kemper and William H Sanders editors Computer Performance Evaluation TOOLS volume 2794 of Lecture Notes in Computer Science pages 218 236 Springer 2003 Peter Buchholz and Jan Kriege A Heuristic Approach for Fitting MAPs to Mo ments and Joint Moments In Proc of the 6th International Conference on Quan titative Evaluation of SysTems QEST 2009 pages 53 62 Los Alamitos CA USA 2009 IEEE Computer Society Marne C Cario and Barry L Nelson Autoregressive To Anything Time Series Input Processes for Simulation Operations Research Letters 19 2 51 58 1996 Marne C Cario and Barry L Nelson Numerical Methods for Fitting and Sim ulating Autoregressive To Anything
51. rithm will use this MAP as initial solution and improve its likelihood The tool specific properties are shown in Fig 30 ANA Edit Job MAP_EM Conv e Format lt eps gt Result 0 005 0 005 v Enable vi Visible lteration Format lt iter gt Result 20 20 v Enable vi Visible Map Dims Format lt dim gt Result 3 3 vi Enable vi Visible Alpha Format lt alpha gt Result 0 5 0 5 vi Enable vi Visible Seed Format lt seed gt Result 123 123 vi Enable vi Visible OK Last Values Defaults Cancel Figure 30 Property window of MAP_EM e Input Trace MAP optional e Output MAP e Parameters 31 Conv Stops MAP Dims determines the order of the MAP 1 e the number of states Alpha is a factor used for the randomisation technique Seed of the random number generator used to generate the random initial MAP MAP MOEA MAP_MOEA 20 combines an EM algorithm for PH distribution fit ting with a multi objective evolutionary algorithm to fit a MAP Its parameters are shown in Fig 31 e Input Trace e Output MAP e Parameters Number of Generation determines how many generations the evo e and Iteration are used to specify the stopping criterion for the algorithm If either the improvement of the algorithm is less than e or the specified number of iterations have been reached the algorithm ANA Edit Job MAP_MOEA Number of Generation
52. rocesses are referred to 16 17 for PH distributions and MAPs to 15 for CAPP processes to 10 for ARTA processes and to 7 for ARIMA models For a general overview of ProFiDo s key features and the motivation that led to the development of ProFiDo the reader is referred to 1 4 Technical documentation on the different file formats that ProFiDo uses for its configuration for saving workflows and for the description of stochastic processes can be found in 2 3 6 This document is structured as follows Installation instructions for ProFiDo are given in Sect 2 Sect 3 contains an introductory tutorial and Sect 4 explains the use of ProFiDo in detail 2 Installation The latest version of ProFiDo can be obtained as a zip archive from http 1s4 www cs tu dortmund de profido The archive contains the GUI ProFiDo and a collection of tools for fitting stochas tic processes and result visualisation The GUI is released under the GPL see Sect 2 1 and the sources are available on the website as well The included tools are currently only available as binaries 2 1 License ProFiDo is free software released under the terms of the GNU General Public Li cense version 2 as published by the Free Software Foundation ProFiDo is distributed in the hope that it will be useful but WITHOUT ANY WARRANTY without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE See the GNU General Public License for more det
53. rov Smirnov test 37 LXTEX table 8 38 latex 39 LBL TCP 3 6 least squares 30 33 Likelihood 28 Load Workflow 23 lognormal 28 MAP 3 MAP_EM 31 MAP_MOEA 32 Matrix Exponential Process 34 maximum likelihood estimate MLE 28 Menu Bar 6 Edit 23 25 File 23 Help 23 26 Settings 23 25 Workflow 23 24 MEP 34 MEPfit 34 Modgen 37 mom 38 39 moments 30 38 Momfit 30 Node allowed connections 19 Connecting 19 Input 19 Job 19 Output 19 Queue 39 non linear optimisation algorithm 30 normal 28 OMNeT 39 order key 20 Output File 18 Formats 18 Output file 8 Output node 8 18 19 parameter description 19 enabled 19 visible 19 PDF 38 Pearson s Chi Squared test 37 perform_tests sh 43 PH 3 Plot 8 38 ac 38 39 at edges 15 cdf 38 39 intermediate results 15 pdf 38 39 PNG 25 38 points Auto Position 20 tooltip 20 Portable Document Format 38 48 Portable Network Graphics 38 PostScript 38 Print 8 ProFiDo acronym 3 command line parameters 42 directory software 4 gpl 2 0 txt 4 installation 3 License 3 profido jar 4 profido sh 4 6 recommended installation directory 4 software xml 4 starting 4 6 system requirements 3 workflow 18 property window 8 PS 38 question mark 19 Queue 39 Quit 23 R 4 35 randomisation technique 32 33 Redo 25 result directory 22 Save Screenshot 25 Save Workflow 23 Scale on resize 25 s
54. sible Sim Time im Format lt sim time gt Result 175000s f 175000s M Enable Visible OK Last Values Defaults Cancel Figure 41 Property window of Queue e Parameters Service Time The service time of the server This value may be either deterministic or drawn from a probability distribution See sections 3 6 8 and 20 from 19 for the available functions and probability distributions Note that the expression must be enclosed in quotation marks and that OMNeT expects a unit of measurement e g exponential 1 0s draws the service time from an exponential distribution with a mean of 1 second Note that the interarrival times used as input for the queue are in seconds Buffer The maximum length of the queue Sim Time The simulation time including a unit of measurement e g L79S0008s 40 A Developing ProFiDo In the following we provide a brief overview of several approaches that can be used to extend ProFiDo and introduce tools helping in that process Sect A 1 provides an overview how additional command line tools can be integrated using the XML based configuration file In Sect A 2 a small test suite consisting of some extensions to the source code and several bash scripts are described A 1 Integrating additional tools ProFiDo provides a flexible environment that can be easily adapted to incorporate ad ditional command line tools cf 1 The GUI is able to handle any kind of
55. the input filename lt seee a a a a a 7 Positioning the input node a a 8 Inserting a job node 4 dim dE oh ee ot Sew eee eB 9 Example of a parameter window ofajobnode 9 Canvas with several job nodes 02 2 004 10 Adding an output file 2 2 0 0 0 000000 10 Selecting the type of output 2 2 2 0 0 02020008 1 Positioning the output node 0004 11 Connecting with destination node 12 Result after connecting two nodes 0 4 13 Example of a complete workflow 14 Plot of autocorrelations for the workflow of Fig 13 15 Plotter funcuonality s 20 2 4 an a e dd eS 16 Property window of the Edge View Plot 16 Property window of the Edge View Print 17 Complete workflow after positioning points 20 PIOMErLUNCHONAIIEY aens hae AME Ahem Wee AL ee e eee e da 21 Property window of the Edge View Plot 21 Property window of the Edge View Print 22 Pile Wen ds bee oe ee RAE SRS OS Ae rs 23 Worktow Meni paid ass oh ee ea Oe eRe eS 24 EII a Be ee SS eS a Bee eS 2 25 Sene S Men noo Gk a oe ge oe a i an Nd ta os See che hy eee ae OH is 25 Help Mente oa o e Be ees Be ene ZG Oe we eG 26 Property window of DistFit 28 Property window of G FIT eee 29 Property window of Momfit
56. visualised 1 Starting ProFiDo by calling profido sh gives a window with an empty can vas as shown in Fig 1 6006 ProFiDo File Workflow Edit Settings Help Figure 1 Empty Workflow 2 Describing a workflow means essentially to describe a graph with edges and nodes So let us start with the first node of the graph describing an input file Input files are usually trace files that contain observations from a real system and that we want to fit with a stochastic process For selecting an input file click on Workflow in the menu bar and select Add Input File see Fig 2 This will open a dialog window where one can select the input file see Fig 3 After that the node which is associated with the input file is placed with a click into the canvas window see Fig 4 For our example workflow we opened the trace LBL TCP 3 22 from the Internet Traffic Archive 25 that was reduced to contain only the interarrival times This version of the trace is also available from ProFiDo s website see Sect 2 File WOO Edit Settings Help Add Job Add Output File Change Workflow Name Set Working Directory Export Workflow as Script Export Workflow as Script with Report Save Screenshot Figure 2 Adding an input file node epa trace B matrix_exp10000 trace B pAug89 trace Dateiformat Figure 3 Selecting the input filename 600 ProFiDo File Workflow Edit Settings Help
57. window similar to the parameters window of the job node Plot see Fig 16 After confirming the values with a click on the OK button the characteristics of the selected model are displayed in a new window In addition to the graphical view a text based display of intermediate results is also possible To display textual informations select the option Print Details see Figs 15 and 17 15 File Workflow Edit Settings Help Joint Moments 5 PrOFIVO Datcn JMomfit Delete Connection Edit Connection Add EdgePoint Delete EdgePoint v Auto Position Show XML File Plot Deta Print Details Figure 15 Plotter functionality Start Format lt start gt Result 0 po mae End Format lt end gt Result 10 Y Enable Type Format lt Type gt Result plot cdf Enable Figure 16 Property window of the Edge View Plot 16 600 Setup edge view Print Format Format lt Format gt Result format txt txt M Enable Relative Format lt relative gt Result false false Y Enable Start Format lt start gt Result 0 0 Y Enable End Format lt end gt Result 10 10 v Enable Type Format lt Type gt Result plot cdf cdf y Y Enable OK Last Values Defaults Cancel Figure 17 Property window of the Edge View Print 4 Working with ProFiDo In the following we will explain in detail how Pro
58. workflow and creates a versatile PDF report file con taining all properties of the workflow and its intermediate as well as final results The PDF report file is generated from IAIFX files generated by ProFiDo which can be used as a basis for further documentation The TEX files are located in directory Results Report Files Table 1 Joint moments for the workflow from Fig 13 Ibl3_dat MAP3_JM MAP3_AC 1 30 1 29 1 30 17 42 17 65 15 97 890 5 890 5 673 9 107474 105484 70378 5 5 21417760 22268852 14002609 The execution of the example workflow will result in two files as specified by the 13 File Workflow Edit Settings Help 3 F kad o E o E Figure 13 Example of a complete workflow 14 Autocorrelations for lags 1 10 Ib13_ dat MAP3_AC MAP3_JM autocorrelation Figure 14 Plot of autocorrelations for the workflow of Fig 13 output nodes in Fig 13 The plot with the autocorrelations is shown in Fig 14 Table 1 shows the joint moments created by the Print node Observe that the curves in Fig 14 and the columns in Table 1 are labeled with MAP 3_AC and MAP 3_JM which corresponds to the file names we entered at the incoming arcs of Plot and Print in Fig 13 Once the workflow has been executed the Edge View functionality can be used to display intermediate results In order to do this right click on the edge and select the option Plot Details see Fig 15 The menu item Plot Details opensa
Download Pdf Manuals
Related Search