cluster de calcul parallèle linux
Contents
1. ethi ethi ethi ethi in out in out in out in out 2 018 EX 2 015 E 2 013 NZD 2 018 EXT 1 I 1 i 1 node4 nodes node node bi This first application is launched automatically by Moniteos at startup You can access it later by clicking on the first icon at the bottom of the page This monitoring applet gives you a synthetic view of the cluster node by node Each node has four indicators CPU Memory Swap and Network state If you select one or more nodes by clicking on the screen icon and then right clicking on one of them you can access statistics of their utilization from one hour to one month Note that if N A is displayed the considered nodes could have problems verify these nodes state Previous Next Return to index 5 4 Textual monitoring d Mozilla Moniteos Cluster monitoring and configuration tool Fichier Edition Affichage Aller Communicator Aide 7 g ak 2 til c a EN Pr c dent suivant Recharger Page d accueil Rechercher Mozilla Imprimer S cur B Signets Adresse http 192 168 0 150 noniteos Infos connexes 4 Red Hat Network 4 Support 4 Shop 4 Products 4 Training al ncos High Performance Linux Solutions bMoniteos EE NUNC ee node3 node4 M bMoniteos Parallel Command Monitoring lt 1002. PBalineoS Job Inter Help i He em ES The textual monitoring application is accessed by clicking on the icon bMoniteos This application allows you to control the cluster state as a table each column describes a particular state of the cluster We precise here what mean the columns labels CPUx the mean rate of utilization of the cpu as a mean number of processes in running state per minute LAS5 mean number of processes in running state for five minutes LA10 mean number of processes in running state for ten minutes LATOT total number of processes in running state on the system MEMUSED amount of used memory on the system MEMFREE amount of free memory on the system MEMSHAR amount of shared memory on the system included in MEMUSED MEMBUF amount of buffered memory on the system included in MEMUSED MEMCACH amount of cache memory used on the system included in MEMUSED SWAPUSED amount of swap used SWAPSIZE amount of swap space on the system IFXNAME name of the x network interface IFxSEND number of bytes sent by this interface per second IFxRECV number of bytes received by this interface per second IFxCOLL number of bytes per second which collided Previous Next Return to index 5 5 The Parallel Command tool Login Execution Result H master0 PID TTY TIME CMD m er0 1154 7 00 00 00 moniteos r PID TTY TIME CMD nodel 7327 00 00 00 moniteos
3. alineos ax High Performance Linux Solutions cluster de calcul parall le linux manuel d utilisation ALINEOS SA Soci t Anonyme au capital de 45 734 70 RCS Montereau B 430 100 628 APE 725 Z 14 bis rue du Mar chal Foch 77780 BOURRON MARLOTTE T l phone 01 64 78 57 65 T l copie 01 64 78 57 66 E mail info alineos com Web www alineos com INDEX 1 CONFIGURATION LOGICIELLE DU CLUSTER 2 PROCEDURE DE DEMARRAGE ET ARRET DU CLUSTER 3 EXPLOITATION DU CLUSTER 4 EXPLOITATION LOGICIELLE 5 MONITEOS v2 6 6 CHECK LIST D INSTALLATION CLUSTER 1 1 CONFIGURATION LOGICIELLE DU CLUSTER SERVEUR S Distribution Linux Red Hat 7 3 Kernel 2 4 18 ou sup image de la distribution opt RH 7 3 nom de s machine s master n domaine com n 0 1 par d faut master0 clustal com etho 192 168 1 15 n netmask 255 255 255 0 r seau 192 168 1 0 diffusion 192 168 1 255 eth1 netmask r seau diffusion partitions des disques durs boot 100 MO 10 GO home gt swap RAM x 2 compte root mot de passe compte admin mot de passe MPI CH 1 2 4 usr local mpich 1 2 4 LamMPI 6 5 7 usr local lam mpi OpenPBS 2 3 16 usr local OpenPBS 2 3 16 ATLAS 3 4 1 usr local ATLAS LAPACK 3 0 usr local LAPACK MPIBLACS 1 1 u
4. node PID TTY TIME CMD node 7407 00 00 00 moniteos Parallel Update Login Command Nodes Command Shortouts Topnode Killnode kd User Passud Clear Status KillRll View All Node Reboot Shutdown The Parallel Command application is accessed by clicking on the third icon This tool allows you to select a set of nodes from one node to all and to type in a command that will be executed by this set of nodes In particular if you need to execute this command under a particular login fill in the two fields Login and Password and then click on Update Login just below You can then type your command in the Command field and click on Execute to begin execution You can see the output of your command either for the whole set of node or node by node Note that any command you type must not prompt user for data input while executing The Status button displays in the main window the state of execution of the command node by node The KillAII button stops commands on each node In addition 5 other buttons were added corresponding to perform programmed actions the Topnode button displays for the selected nodes the 2 jobs having the most cpu weight the Killnode button allows the super user to kill a named job on the selected nodes the MAJ User Passwd allows the super user to propagate the user authentication from the server to the selected nodes Please remember to create an
5. np 2 properties blue cluster Note Qstat button and PBS Nodes button are also available from the access page of Pbalineos Previous Next Return to index 5 5 Parallel jobs PBalineoS and MPI To use the full features of mpi ch implementation to OpenPBS via PBalineoS you only need to give following arguements to your command line for mpi ch over IP machinefile confile np nnodes and for mpi ch over GM gm f confile np nnodes Previous Return to index 5 8 Interactive Jobs CPU Mem CPU Mem master 0 CPU CPU Mem CPU Mem Load script ra IH 1 node E fusr local bin x pympoy S i home admi 1480 0640 FT i home admin pov skyvase pov 80 0640 vi nodez d a0 300 q9 mv2 0 bi000 nw32 nh32 nt8 L home admin pov pympov3_1g_1 povray31 inelude H node3 D Umaster0 1 I I I 1 1 f Submit The Interactive Job Submission application is accessed by clicking on the fifth icon This tool is a web interface which allows you to submit jobs to a selected node with respect to its cpu and memory load To submit a job you only need to do following ajob name optional e select a script file by clicking the Load Script button You will browse in your directories tree and select a file containing the appropriate instructions for the
6. noden du cluster qui doivent tre arr t es ne pas s lectionner le serveur 3 lancer la commande parall le shutdown 4 surles machines SMP arr t lectrique manuel des machines option 2 arr t manuel 1 arr t des n uds noden n 0 7 du cluster depuis le serveur su rlogin noden n 0 7 halt ou shutdown h now 2 arr t du serveur su shutdown h now 3 surles machines SMP arr t lectrique manuel des machines d brancher le s switch s 3 EXPLOITATION DU CLUSTER 3 1 RECOMMANDATIONS GENERALES Privil gier la connexion utilisateur la connexion root Se connecter sur master0 sur un compte utilisateur La connexion aux n uds de calcul doit tre r alis e sans authentification rhosts en droits 600 dans la racine de l utilisateur Sauvegardes a minima il est recommand d effectuer les sauvegardes des r pertoires utilisateurs c est dire de home sur le serveur 3 2 MANAGEMENT DU CLUSTER moniteos monitoring statistiques et commandes parall les se connecter via un navigateur web l adresse http master0 moniteos index html Se reporter la documentation sp cifique pour l installation et l utilisation de moniteos 3 3 DEMARRAGE D UN N UD DE CALCUL APRES LE DEMARRAGE DU SERVEUR Pour remonter les syst mes de fichiers export s nfs par le serveur passer su sur le n ud de calcul et lancer bash mount a 3 4 DEMARRAGE D UN N UD DE CALCUL SANS SERVEUR Si le ser
7. Bourron Marlotte France http www alineos com 6 CHECK LIST D INSTALLATION CLUSTER V rification des partitions Sur le frontal df Sur les noeuds Jexecnode df k t 7 V rifier le montage de usr local et home V rifier le swap free grep Swap V rification de versions noyau Sur le frontal uname a Sur les n uds execnode uname a t 7 V rification des taux d acc s disques Sur le frontal hdparm dev hda Sur les noeuds Jexecnode hdparm tT dev hda t 7 V rification de la configuration IP V rifier la configuration IP cluster Param trer l IP externe DNS Gateway et nom de machine etc hosts V rification des taux d acc s r seau MPI Script de lancement de l ex cutable d exemple de mpich systest dans home admin mpisystest sh e Moniteos Verification version Logo High Performance Linux Solutions Monitoring V rifier que tous les noeuds sont bien en ligne BMoniteos V rifier que les donn es sont bien renseign es pour tous les noeuds Parallel Command Lancer une commande sur plusieurs noeuds exemple echo Pbalineos qstat PBS nodes lancement du calcul de pi depuis le script home admin mpicpi pbalineos sh OPEN PBS Lancer trois fen tres xterm Dans le premier taper qsub I l nodes 2 ppn 2 Dans le second taper qsub I 1 nodes lt X gt ppn 2 avec X de sorte que X 2 Nombre total de n ud du clust
8. appropriate rhosts thereafter Make Boot disk it s only to make a boot disk Install Node please refer to the PRO version documentation Previous Next Return to index 5 5 The PBalineoS batch system tool i Mozilla Moniteos Cluster monitoring and configuration tool Fichier Edition Affichage Aller Communicator Aide 7 2 a 2 eu aj EN Pr c dent suivant Recharger Page d accueil Rechercher Mozilla Imprimer S cur wu Signets A Adresse fhttp 192 168 0 150 Infos connexes 4 Red Hat Network 4 Support 4 Shop 4 Products G Training alncos High Ferformance Linux Solutions v Sequential job 1 w Parallel job ppn 2 Nodes You can change your Login above if necessary Login Monitoring bMoniteos Parallel PBalineoS Job Inter Help Command He Jel AP EJ we Fy hitp vwew alineos com The PBalineoS application is accessed by clicking on the fourth icon This tool is a web interface to OpenPBS It allows you to submit jobs to the PBS server according to the defined PBS attributes of corresponding nodes and queues First you will have to choose what type of jobs you want to submit to 1 sequential or parallel 2 on ip based network or gm Myrinet based network 3 with 1 processor per node or 2 processors Then a new window opens Previous Next Return to index Jo
9. b name 0456 tus Queue to submit job to Ini Number of nodes to use 2a Maximum time HH MM SS 00 00 00 no time limit 1 00 00 d Merge STDERR to STDOUT Send message when job J Rborts J Ends J Starts admin master0 clustal coni LE File Staging data files only executable automaticaly stage Stagein From here To there Stageout From here SO To there Load Script bin sh usr local mpich 1 2 1 7 bin mpirun ch gm gm use shmem np 4 home admin C HMpi datpf home admin C Mpi Test s par 130 gt home admin result 4 130 Create Script This script file was automatically generated by PBAlineoS bin sh S bin sh SPBS N 0456_ 05 PBS q One PBS 1 walltime 1 00 00 nodes 2 ppn 2 PES M admin master0 clustal com bin bash nodes cat PBS_NODEFILE nnodes wc PBS_NODEFILE awk print 1 confile ip PBS_JOBID conf touch confile je init for i in nodes do if j i then Submit PEs Nodes Previous To submit a job you need to give the following informations e ajob name optional the name of the selected queue the number of nodes you need the maximum time of the job 00 00 00 means no limit do you want to merge the stdout with the stderror Yes or No anemail address to send m
10. des you selected There is no limitation on the number of nodes Next Return to index 5 2 Accessing Moniteos Mozilla Moniteos Cluster monitoring and configuration tool x Fichier Edition Affichage Aller Communicator Aide y a a 2 cl SEN Pr c dent Suivant Recharger Page d accueil Rechercher Mozilla Imprimer S cur wu Signets A Adresse http 192 168 0 150 noniteos I Infos connexes 4 Red Hat Network 4 Support 4 Shop 4 Products 4 Training alncos High Ferformance Linux Solutions User admir Monitoring bMoniteos Parallel PBalineoS Job Inter Help Command m 10 http www alineos com ER SP The Moniteos interface is accessed by a web browser you have to connect to the master node of the cluster and then assuming its name is master0 browse to http master0 moniteos This will lead you to Moniteos main page Note Only the help icon is available without log in Previous Next Return to index 5 3 Graphic Monitoring node 0 node 1 node2 node3 I i Le M 6o o Cpu 1 Cpu orne Ree SSR ary FRA SW Memory Memory Memory Memory nn 2252222 Swap Swap 1 Swap Swap Net Net Net Net etho etho etho etho in out in out in out in out 0 59KF EMEIS 0 26KF 0 5 KES _ 2 KES
11. e dans usr local LAPACK 4 4 BLACS BLACS MPI version 1 0 BLACS PVM version 1 0 installation dans usr local BLACS librairies BLACS MPI dans usr local BLACS lib librairie BLACS PVM dans usr local BLACS lib 4 5 SCALAPACK version 1 7 installation dans usr local SCALAPACK librairies SCALAPACK MPI PVM dans usr local SCALAPACK 4 6 OPEN PBS version 2 3 16 installation dans usr local OpenPBS 2 3 16 param trage des machines du cluster dans usr spool PBS server priv nodes 4 7 LAM MPI version 6 5 7 compilation dans usr local lam 6 5 7 installation dans usr local lam mpi param trage des machines du cluster dans usr local lam 6 5 7 etc lam bhost def et usr local lam mpi etc lam bhost def 4 8 PVM version 3 4 4 installation dans usr local pvm3 param trage des machines du cluster dans home admin pvmhostfile 5 MONITEOS v2 6 5 1 Introduction 5 2 Accessing Moniteos 5 3 Graphic Monitoring 5 4 bMoniteos Textual monitoring 5 5 The Parallel Command tool 5 6 The PBalineoS batch system tool 5 7 PBalineoS and mpi ch 5 8 Interactive Jobs 5 9 License agreement 5 1 Introduction Moniteos is a complete cluster administration tool It allows you to watch in real time the cluster load as a graphic or textual report you can also obtain statistics on the utilization of the cluster node by node You can always through the same web interface execute instructions that will be treated by the no
12. er 1 Dans le troisi me taper qstat n et v rifier l tat des jobs le second doit tre mis en attente ATLAS Librairies disponibles dans usr local ATLAS lib lt ARCH gt Installation de gcc 2 95 3 permettant une meilleure optimisation Utilisable par source home admin gcc 2 95 3 LAPACK Librairie dans usr local LAPACK BLACS Librairies MPI et PVM dans usr local BLACS LIB SCALAPACK Librairies MPI et PVM dans usr local SCALAPACK PVM Configuration de nouveaux comptes 4 lignes rajouter au fichier bashrc Cf home admin bashrc e Test depuis le nodeO dans home admin pvm pvmhostfile puis au promt pvm gt conf renvoie la configuration du cluster pvm gt spawn gt hello gt halt LAM T Configuration de nouveaux comptes 3 lignes rajouter au fichier bashrc Cf home admin bashrc Test depuis le nodeO dans home admin recon test la configuration pour LAM lamboot s lamhostfile lance LAM avec la configuration du fichier lamnodes affiche la configuration home admin TestLam sh script qui lance deux programmes d exemples LAM ring et cpi calcul de pi lamhalt Deux fichiers r sultats se trouvent dans home admin e lampitest txt e lamringtest txt
13. essages when jobs starts ends or aborts the filenames of stagein and stageout files and then select a script file by clicking the Load Script button Previous Next Return to index admin20011227 1 032 dun mpi dun mpi anm dun pum null20011227 16 e 3 null20011227 16 03 File name dun mpi an Directorv name home admin ok Unsigned Java Applet Window You will browse in your directories tree and select a file containing the appropriate instructions for the job The script is then loaded in the corresponding applet s window In this window you always can modify the pre loaded instructions After this step you will create the PBS script by clicking the Create Script button and all the instructions then appear in the second applet s window Once again you can modify in this window all the instructions including PBS directives Finally submit your job by clicking the Submit button Previous Next Return to index K Queue Status When running PBS you at every time can select a queue and then click on the Qstat button to display the status of the selected queue You can also click on the PBS Nodes button to display the status of all the node PBS Nodes job exclusive cluster jobs 0 554 master0 clustal com 1 554 master0 clustal com state job exclusive np 2 ntype cluster jobs 0 554 master0 clustal com 1 554 master0 clustal com state free
14. job The script is then loaded in the corresponding applet s window In this window you always can modify the pre loaded instructions e After this step you will choose the node you want to submit the job to Finally submit your job by clicking the Submit button Previous Return to index 5 9 Moniteos License Agreement NOTICE PLEASE READ CAREFULLY THIS DOCUMENT BEFORE USING THE SOFTWARE BY USING THE SOFTWARE YOU ARE AGREEING TO BE BOUND BY THE TERMS OF THIS LICENCE IF YOU DO NOT AGREE TO THE TERMS OF THIS LICENCE PLEASE DO NOT USE THE SOFTWARE You as a Licensee under this Agreement are hereby granted a limited irrevocable nontransferable and nonexclusive license to use the Software subject to the restrictions and other terms within That use must be only by the Licensee Any complete or partial reproduction or distribution of the software or its source code is strictly prohibited Licensee may not rent lease loan electronically transfer the Software and its source code to others The Licensee may modify the source code for his own use YOU ACKNOWLEDGE THAT YOU HAVE READ THIS AGREEMENT AND AGREE TO BE BOUND BY ITS TERMS YOU FURTHER AGREE THAT IT IS THE COMPLETE AND EXCLUSIVE STATEMENT OF AGREEMENT BETWEEN US WHICH SUPERSEDES ANY PRIOR AGREEMENT ORAL OR WRITTEN ANY PROPOSAL AND ANY OTHER COMMUNICATIONS BETWEEN US RELATING TO THE SUBJECT MATTER OF THIS AGREEMENT ALINEOS 14 bis rue du Marechal Foch 77780
15. sr local BLACS PVMBLACS 1 1 usr local BLACS PVM 3 4 4 usr local pvm3 SCALAPACK 1 7 usr local SCALAPACK 1 2 CONFIGURATION LOGICIELLE DU CLUSTER N UDS DE CALCUL Distribution Linux Red Hat 7 3 Kernel 2 4 18 ou sup N uds de calcul noms node i domaine com par d faut master0 clustal com etho 192 168 1 1 0 i netmask 255 255 255 0 r seau 192 168 1 0 diffusion 192 168 1 255 partitions du disque dur boot 100 MO 3 GO tmp gt swap RAM x 2 montage nfs master0 home sur home master0 usr local sur usr local floppy compte root mot de passe compte admin mot de passe 1 3 SWITCH FAST ETHERNET pour HP et 3Com manageables IP 192 168 1 98 login login constructeur mot de passe mot de passe constructeur interface web http 192 168 1 98 2 PROCEDURE DE DEMARRAGE ET ARRET DU CLUSTER La proc dure de d marrage du cluster est la suivante 1 v rifier l alimentation lectrique 2 brancher le switch Fast Ethernet 3 d marrer le s serveur s 4 attendre l cran de login du serveur 5 d marrer les nceuds de calcul 6 travailler La proc dure d arr t du cluster est la suivante option 1 arr t par moniteos 1 passer su dans la fen tre des commandes parall les de moniteos 2 s lectionner les n uds
16. veur n est pas d marr le montage nfs de home n est pas r alis et les comptes utilisateurs NIS ne sont pas op rationnels S il s av re n cessaire de travailler sur un n ud en station isol e avec des comptes utilisateurs sans pour autant perdre la configuration cluster proc der de la mani re suivante brancher un cran et un clavier minimum requis se connecter en tant que root ou admin puis su 1 arr ter le d mon ypbind 2 diter etc fstab commenter les lignes ins rer en d but de ligne master0 home master0 usr local Il est alors possible de travailler avec une station pleinement op rationnelle nb le compte admin est cr sur chaque n ud 4 EXPLOITATION LOGICIELLE 4 1 MPICH version 1 2 4 compilation dans usr local mpich 1 2 4 installation dans usr local mpich si un seul compilateur installation dans usr local mpich gcc et mpich autre compilateur si n cessaire param trage des machines du cluster usr local mpich 1 2 4 util machines machines LINUX et usr local mpich compilateur share machines LINUX renseigner avec les noeuds de calcul important hostname sur machine SMP compilation avec l option with comm shared 4 2 ATLAS version 3 4 1 installation dans usr local ATLAS librairies BLAS optimis es dans usr local ATLAS lib Linux PASSE2 2 pour la configuration nceud 4 3 LAPACK version 3 0 installation dans usr local LAPACK librairie LAPACK optimis
Download Pdf Manuals
Related Search