PAR Technologies V5 manual Using ParaStation process pinning, Using memory binding

Page 31

Using ParaStation process pinning

To list, sort and filter all the collected information, the command psaccview is available.

See psaccounter(8) and psaccview(8) for details.

5.19. Using ParaStation process pinning

ParaStation is able to pin down compute tasks to particular cores. This will avoid 'hoping' processes between different cores or CPUs during runtime, controlled by the OS scheduler.

While placing tasks to particular nodes, ParaStation will also decide which CPU-slot (= virtual core) on this node will be used. The physical core assigned to this CPU-slot will be calculated using a mapping list defined in the configuration file.

Process pinning may be enabled or disabled globally or on a per node basis. Refer to pinProcs and CPUmap entries in parastation.conf and the set pinprocs and set cpumap directives of psiadmin for details.

See also parastation.conf(5) and psiadmin(1) for more information.

5.20. Using memory binding

Beside pinning down compute tasks to particular cores, ParaStation is also able to use memory binding techniques on NUMA based systems. This will give hints to the memory management subsystem of the operating system to select 'nearest' memory, if available.

Memory binding may be enabled or disabled globally or on a per node basis. Refer to the bindMem entry in parastation.conf and set bindmem directive of psiadmin for details.

See also parastation.conf(5) and psiadmin(1) for more information.

5.21. Spawning processes belonging to all groups

By default, newly created processes only belong to the primary group for the user ID as definined on the spawning node. To add a process to all groups a user belongs to on the current node, enable this flag using the supplementaryGroups directive or set the configuration flag supplGrps in parastation.conf to true.

Enabling this behaviour may trigger extensive network traffic, depending on how the user authentication is configured on the nodes. E.g., using LDAP will open a connection to the LDAP server.

See also parastation.conf(5) and psiadmin(1) for more information.

5.22. Changing the default ports for psid(8)

By default, the ParaStation daemon psid(8) uses the port 888 for TCP connections. To change this port, modify the files /etc/services and /etc/xinet.d/psidstarter.

Add the following line to /etc/services:

psid

888/tcp

# ParaStation Daemon Start Port

ParaStation5 Administrator's Guide

27

Image 31
Contents Administrators Guide Info@par-tec.com ParaStation5 Administrators GuideTable of Contents Problem different groups of nodes are seen as up or down History of ParaStation IntroductionAbout this document Kernel modules Technical overviewRuntime daemon LibrariesLicense Prerequisites InstallationHardware Software Directory structureKernel version Mpi2, mpi2-intel, mpi2-pgi, mpi2-psc Installation via RPM packagesGetting the ParaStation5 RPM packages ManCompiling the ParaStation5 packages from source Installing the RPMsFile Version Etc/init.d/xinetd reload Installing the documentationParaStation entries # rpm -Uv psmpi2.5.0.0-1.i586.rpm Installing MPIFurther steps # rpm -Uv psdoc-5.0.0-1.noarch.rpmUninstalling ParaStation5 ParaStation5 Administrators Guide Define Number of nodes ConfigurationConfiguration of the ParaStation system Copy template# /opt/parastation/bin/testconfig Enable optimized network driversHostname id HWType runJob starter accounter Testing the installation # /opt/parastation/bin/testnodes -np nodes # /opt/parastation/bin/psiadmin -s -c listParaStation5 pscom communication library Insight ParaStation5Directory /proc/sys/ps4/state # echo 10 /proc/sys/ps4/state/ResendTimeout# cat /proc/sys/ps4/state/connections Directory /proc/sys/ps4/local Controlling process placementUsing non-ParaStationapplications Using the ParaStation5 queuing facilityExporting environment variables for a task Export LDPRELOAD=/opt/parastation/lib64/libp4tcp.so Controlling ParaStation5 communication pathsExport PSPLIB=/opt/parastation/lib64/libpscomopenib.so Authentication within ParaStation5Pspshm or Pspsharedmem PSPP4S or PSPP4SOCKNodes and CPUs Homogeneous user ID spaceSingle system view Parallel shell toolPSIRARGPRE0=/some/path/env2tok Integration with AFSIntegrating external queuing systems Tok2envMulticasts Route add -net 224.0.0.0 netmask 240.0.0.0 dev ethX Using ParaStation accountingCopying files in parallel # UseMCastSpawning processes belonging to all groups Using ParaStation process pinningUsing memory binding Changing the default ports for psid8Port Problem psiadmin returns error TroubleshootingProblem node shown as down Problem cannot start process on frontend Problem cannot start parallel taskProblem bad performance Problem different groups of nodes are seen as up or downProblem psid does not startup, reports port in use Problem pssh failsProblem processes cannot access files on remote nodes Reference Pages ParaStation5 Administrators Guide Parameters InstallDir inst-dir , InstallationDir inst-dirParastation.conf DescriptionStatusscript SetupscriptStartscript StopscriptElan P4sockOpenib MvapiNrOfNodes num Accounter$GENERATE 1-96 node$0,2 $0 Node node17 16 HWType ethernet p4sock starter yes runJobs noMCastGroup group-num SelectTime timeDeadInterval num LogLevel numMemLock size Core sizeCPUTime time DataSize sizeProc CPUmap map Processes maxprocsRdpResendTimeout ms RdpTimeout msStatusTimeout ms RdpClosedTimeout msSee also ErrorsParaStation5 Administrators Guide Synopsis PsiadminOptions Extended description Standard ErrorStandard Input Standard OutputAll ExitAllproc cnt count Load Count hw hwDown HardwareSummary max max RdpQuit Maxproc nodes Accounters nodesUser nodes Group nodesNodesSort nodes Master nodesFreeOnSuspend nodes HandleOldBins nodesRlrss nodes Cpumap nodesRdpResendTimeout nodes RdpTimeout nodesStatusTimeout nodes RdpClosedTimeout nodesRestart nodes Resolve nodesPsiddebug mask nodes Selecttime time nodesPattern Name Description HandleOldBins 0 1 nodes Rdpmaxretrans val nodes RdpResendTimeout ms nodes RdpTimeout ms nodesStatusTimeout ms nodes RdpClosedTimeout ms nodesVerbose FilesQuiet NormalPsid Debug=level Configfile=fileLogfile=file ? , --usage Show a help message TestconfigFilename NumParaStation5 Administrators Guide Map TestnodesNp num Cnt numParaStation5 Administrators Guide Testpse -npnum TestpseParaStation5 Administrators Guide ?,--help P4statSock NetParaStation5 Administrators Guide Add P4tcpDelete ParaStation5 Administrators Guide Pattern Description PsaccounterVar/account/yyyymmdd Accounting files, one per day DumpcoreCoredir=dir ?, --helpPsaccview Ls,--ltotsum Lj,--ljobsLu,--ltotuser Lg,--ltotgroupEnd CpuweightAqtime CputimeInitialization file Mlisten ParaStation5 Administrators Guide Appendix A. Quick Installation Guide # chkconfig -a /etc/init.d/parastation # /opt/parastation/bin/psiadmin psiadmin addTesting Appendix B. ParaStation license Page Page Page Changes to the runtime environment Building and installing ParaStation5 packages# psiadmin -s Appendix C. Upgrading ParaStation4 to ParaStation5Page ARP GlossarySee ParaStation Logger To share a common address space within a node ParaStation5 Administrators Guide