Manuals / HP / Computer Equipment / Software

HP XC System 2.x Software Tuning Applications, Using the Intel Trace Collector/Analyzer, Example

Models: XC System 2.x Software

1 154

Download 154 pages 20.06 Kb

Page 69

5

Tuning Applications

This chapter discusses how to tune applications in the HP XC environment.

5.1 Using the Intel Trace Collector/Analyzer

This section describes how to use the Intel Trace Collector (ITC) and Intel Trace Analyzer (ITA) with HP-MPI on an HP XC system. The Intel Trace Collector/Analyzer were formerly known as VampirTrace and Vampir, respectively. The following topics are discussed in this section:

•Building a Program (Section 5.1.1)

•Running a Program (Section 5.1.2)

•Visualizing Data (Section 5.1.3)

The following Intel Trace Collector/Analyzer documentation is available on the HP XC system after Intel Trace Collector/Analyzer is installed:

•Intel Trace Collector Users Guide — located on the HP XC system at:

<install-path-name>/ITC/doc/Intel_Trace_Collec- tor_Users_Guide.pdf.

•Intel Trace Analyzer Users Guide — located on the HP XC system at:

<install-path-name>/ITA/doc/Intel_Trace_Ana- lyzer_Users_Guide.pdf

5.1.1Building a Program — Intel Trace Collector and HP-MPI

HP-MPI is MPICH compatible if you use the HP-MPI MPICH scripts which are located at: /opt/hpmpi/bin. The HP-MPI MPICH scripts are:

•mpicc is replaced by mpicc.mpich

•mpif77 is replaced by mpif77.mpich

•mpirun is replaced by mpirun.mpich

In summary, mpixx becomes mpixx.mpich.

For further information, refer to the Intel Trace Collector Users Guide. This document is located on the HP XC system at the following location: <install-path-

name>/ITC/doc/Intel_Trace_Collector_Users_Guide.pdf.

Example

For the purposes of this example, the examples directory under /opt/IntelTrace/ITC is copied to the user’s home directory and renamed to examples_directory.

The GNU makefile looks as follows:

CC= mpicc.mpich

F77 = mpif77.mpich CLINKER = mpicc.mpich FLINKER = mpif77.mpich IFLAGS = -I$(VT_ROOT)/include CFLAGS = -g

FFLAGS = -g

LIBS	= -lvtunwind -ldwarf -lnsl -lm -lelf -lpthread

Tuning Applications 5-1

Image 69

HP XC System 2.x Software manual Tuning Applications, Using the Intel Trace Collector/Analyzer, Example

Contents

Hewlett-Packard Company Palo Alto, California June Product VersionPage Contents 5.4 Developing ApplicationsTuning Applications Using SlurmDebugging Applications Using HP-MPI Using LSF6.4 Using HP MlibAdvanced Topics ExamplesGlossary Index Examples Tables FiguresPage About This Document Intended AudienceDocument Organization HP XC Information Linux Administration HandbookQuickSpecs for HP XC System Software HP XC Program Development EnvironmentSupplementary Information For More InformationHP Message Passing Interface HP Mathematical LibraryHttp//supermon.sourceforge.net ManpagesHttp//systemimager.org Http//sourceforge.net/projects/modulesRelated Information Http//linuxvirtualserver.orgHttp//www-unix.mcs.anl.gov/mpi # cd /opt/hptc/config/sbin Typographical ConventionsBold text Ctrl/x HP Encourages Your CommentsPage System Architecture Overview of the User EnvironmentOperating System Node SpecializationFile System Storage and I/OSAN Storage Local StorageFile System Layout System Interconnect NetworkUser Environment CommandsNetwork Address Translation NAT 1 LVSApplication Development Environment Parallel ApplicationsSerial Applications Slurm Run-Time EnvironmentLoad Sharing Facility LSF-HPC How LSF-HPC and Slurm InteractHP-MPI Components, Tools, Compilers, Libraries, and Debuggers8Overview of the User Environment Configuring Your Environment with Modulefiles Using the SystemLVS Login Routing Using ssh to Log2Using the System Supplied Modulefiles Supplied ModulefilesModulefile Sets the HP XC User Environment Viewing Available Modulefiles Modulefiles Automatically Loaded on the SystemViewing Loaded Modulefiles Loading a ModulefileUnloading a Modulefile Automatically Loading a Modulefile at LoginModulefile Conflicts Loading a Modulefile for the Current SessionViewing Modulefile-Specific Help Creating a Modulefile$ module load modules $ man modulefile $ module help totalviewIntroduction Launching and Managing Jobs Quick StartGetting Information About Queues Getting Information About ResourcesLaunching Jobs Getting Information About the System’s PartitionsSubmitting a Serial Job Example 2-1 Submitting a Serial JobSubmitting a Non-MPI Parallel Job Using Slurm Options with the LSF External SchedulerExample 2-2 Submitting a Non-MPI Parallel Job $ bsub -n4 -I srun hostnameSubmitting an MPI Job Example 2-4 Running an MPI Job with LSF$ bsub -n4 Mpirun -srun ./helloworld Submitting a Batch Job or Job Script Example 2-6 Submitting a Job Script$ bsub -I -n4 myjobscript.sh Getting System Help and Information Performing Other Common User Tasks$ man -k keyword $ man sinfoPage Overview Developing ApplicationsStandard Linux Compilers Using CompilersIntel Compilers PGI CompilersChecking Nodes and Partitions Before Running Jobs Setting Debugging OptionsInterrupting a Job Developing Serial ApplicationsDeveloping Parallel Applications Using Mlib in Serial ApplicationsSerial Application Build Environment Building Serial ApplicationsModulefiles Parallel Application Build EnvironmentHP-MPI OpenMPMlib Math Library Quadrics ShmemMPI Library $ mpicc object1.o ... -pthread -o myapp.exePGI Fortran and C/C++ Compilers Intel Fortran and C/C++CompilersGNU C and C++ Compilers GNU Parallel MakeBuilding Parallel Applications Reserved Symbols and NamesCompiling and Linking Non-MPI Applications Compiling and Linking HP-MPI ApplicationsDeveloping Libraries Designing Libraries for XC4000$ mpicc -c -g foo.c Advanced Topics Using the GNU Parallel Make CapabilityExample 3-1 Directory Structure Example 3-2 Recommended Directory Structure$ cd subdir srun -n1 -N1 $MAKE -j4 $ cd subdir srun $MAKEExample Procedure $ make PREFIX=’srun -n1 -N1’ MAKEJ=’-j4’ $ make PREFIX=’srun -n1 -N1 MAKEJ=’-j4’3 I/O Performance Considerations Local Disks on Compute NodesShared File View Private File ViewCommunication Between Nodes Page Debugging Serial Applications Debugging ApplicationsDebugging Parallel Applications TotalViewSetting Up TotalView Debugging with TotalViewSSH and TotalView Using TotalView with LSF-HPC Using TotalView with Slurm$ srun -Nx-A $ mpirun -tv -srun application $ bsub -nx-ext SLURMnodes=x \ -Is /usr/bin/xterm$ totalview Starting TotalView for the First TimeTotalView Preferences Window Preferences window, click on the Launch Strings tab Debugging Applications Debugging an Application $ mpicc -g -o Psimple simple.c -lm$ mpirun -tv -srun -n2 ./Psimple TotalView Process Window Example $ mpirun -srun -n2 Psimple Debugging Running Applications$ scancel --user username Exiting TotalView$ squeue Page Using the Intel Trace Collector/Analyzer Tuning ApplicationsBuilding a Program Intel Trace Collector and HP-MPI ExampleRunning a Program Intel Trace Collector and HP-MPI Visualizing Data Intel Trace Analyzer and HP-MPIExample Running the vtjacobic Example Program Slurm Commands Using SlurmCommand Function IntroductionLaunching Jobs with the srun Command Accessing the Slurm ManpagesSrun Roles and Modes Example 6-1 Simple Launch of a Serial ProgramSrun Modes Srun RolesSrun Run-Mode Options Srun Signal HandlingBatch AllocateSrun Resource-Allocation Options Part --partition=part Cpt --cpus-per-task=cptMinutes --time=minutes Nthreads --threads=nthreadsSrun Control Options Srun I/O Options Mode --input=mode Mode --output=modeMode --error=mode LabelClist --constraint=clist Srun Constraint OptionsMem=size Contiguous=yesnoMincpus=n Vmem=sizeUsing srun with HP-MPI Monitoring Jobs with the squeue CommandUsing srun with LSF Srun Environment VariablesGetting System Information with the sinfo Command Killing Jobs with the scancel CommandJob Accounting Fault ToleranceSecurity $ sinfo -RUsing LSF Introduction to LSF in the HP XC EnvironmentOverview of LSF Nodelist= list-of-nodes Exclude= list-of-nodes Topology Support$ bsub -n 10 -ext SLURMnodes=10exclude=n16 srun myapp $ bsub -n 10 -ext SLURMnodes=10 srun myappHow LSF and Slurm Launch and Manage a Job $ bqueues -l normal grep Jobstarter$ bsub -Is hostname $ bsub -n4 -ext SLURMnodes=4 -o output.out ./myscript How LSF-HPC and Slurm Launch and Manage a JobDifferences Between LSF on HP XC and Standard LSF Determining Available System Resources Determining Execution HostGetting Status of LSF Job Startup and Job ControlGetting Host Load Information Getting Information About LSF-HPC Execution Host NodeChecking LSF System Queues Submitting JobsGetting Information About the lsf Partition $ sinfo -p lsfBsub bsuboptions jobname joboptions Summary of the LSF bsub Command FormatSlurm Arguments Function LSF-SLURM External SchedulerStarting on lsfhost.localdomain n6 Submitting a Job in Parallel Submitting a Serial JobSubmitting an HP-MPI Job Example 7-5 Submitting an Interactive Serial JobSubmitting a Batch Job or Job Script Example 7-6 Submitting an HP-MPI Job$ bsub -n4 -I mpirun -srun ./helloworld Example 7-8 Submitting a Batch Job Script Examples$ bsub -n4 -I ./myscript.sh $ bsub -n4 -ext SLURMnodes=4 -I ./myscript.shSubmitting a Job from a Non-HP XC Host $ bsub -n4 -I ./myscript.sh -n8 -OType=SLINUX64 Getting Job Allocation Information Getting Information About JobsJob Allocation Information for a Running Job $ bsub -R type=SLINUX64 -n4 -I srun hostnameExample 7-13 Using the bjobs Command Short Output Checking Status of a JobJob Allocation Information for a Finished Job Example 7-15 Using the bhist Command Short Output Example 7-14 Using the bjobs Command Long OutputOutput Provided by the bhist Command Viewing a Job’s Historical InformationExample 7-16 Using the bhist Command Long Output Working Interactively Within an LSF-HPC AllocationSubmitting an Interactive Job to Launch the xterm Program Example 7-18 View Your Allocation in Slurm Example 7-17 View Your EnvironmentExample 7-19 View Your Running Job in LSF Example 7-20 View Job Details in LSFExample 7-21 Running Jobs from an xterm Window Submitting an Interactive Job to Launch a ShellExample 7-22 Submitting an Interactive Shell Program $ hostname $ srun hostname $ srun -n2 hostnameLSF Equivalents of Slurm srun Options LSF Equivalents of Slurm srun Options$ srun hostname n1 $ exit Srun Option Description LSF Equivalent24Using LSF Control the parallel job Page Using HP-MPI HP-MPI Directory Structure Setting Environment VariablesCompiling and Running Applications Building and Running an Example ApplicationBuilding and Running helloworld Example Application helloworld$ mpicc -o helloworld $MPIROOT/help/helloworld.c $ $MPIROOT/bin/mpirun -srun -n4 helloworldHP-MPI options allowed with -srun Launching MPI JobsSystem Interconnect Selection Creating Subshells and Launching Jobsteps$ mpirun -srun -n6 -O -N2 -m cyclic ./a.out host1 rank1 $ mpirun -srun -n4 -N2 -O -m cyclic ./a.out host1 rank1$ mpirun -subnet 192.168.1.1 -prot -srun -n4 ./a.out Using LSF and HP-MPISystem Interconnect Support $ /usr/sbin/ifconfig -aMPI Versioning Example 8-5 Allocating 12 Processors on 6 Nodes32-Bit Builds on XC4000 Truncated MessagesAllowing Windows to Use Exclusive Locks $ mpirun -TCP -srun -N8 ./a.out Mpirun Command Options$MPIROOT/bin/mpirun -v -prot -np 2 /path/to/program.x Environment VariablesExport MPIPINPERCENTAGE=30 Export MPIPHYSICALMEMORY=1048576Export MPIPAGEALIGNMEM=1 Export MPIMAXWINDOW=10Export MPIUSELIBELAN=0 Mpich Object CompatibilityExport MPIUSELIBELANUSE=5 $MPIROOT/bin/mpirun.mpich -np 2 ./prog.xAt http//docs.hp.com HP-MPI Documentation and ManpagesHP-MPI Manpage Categories Category Manpages DescriptionAdditional Information, Known Problems, and Work-arounds Intel Compiler Notes Using HP MlibVersion 8 Fortran Compiler Version 7 Fortran CompilerMlib and Module Files HP Mlib for the HP XC6000 PlatformPlatform Support Library SupportUsing Intel Compilers with HP Mlib Modulefiles and MlibCompiling and Linking Licensing HP Mlib for the HP XC4000 PlatformMlib Manpages Linking SuperLUDIST$ pgcc options file 6Using HP Mlib Enabling Remote Execution with OpenSSH Advanced TopicsRunning an X Terminal Session from a Remote Node Determining IP Address of Your Local MachineLogging in to HP XC System Running an X terminal Session Using Slurm$ hostname $ host mymachineRunning an X terminal Session Using LSF $ bsub -n4 -Ip srun -n1 xterm -display$ bjobs Page Examples Building and Running a Serial ApplicationLaunching a Serial Interactive Shell Through LSF Example 1. Two Processors on Any Two Nodes Running LSF Jobs with a Slurm Allocation RequestLaunching a Parallel Interactive Shell Through LSF Example 2. Four Processors on Two Specific NodesView the job $ bsub -Is -n4 -ext SLURMnodes=4 /bin/bash Check the running job’s informationCheck the finished job’s information Submitting a Simple Job Script with LSFShow the environment Display the scriptShow the job allocation Submitting an Interactive Job with LSFShow the Slurm job ID $ bsub -n8 -Ip /bin/shSubmitting an HP-MPI Job with LSF View the running job View the finished job$ bjobs -l $ bsub -n 8 -R ALPHA5 SLINUX64 \ -ext SLURMnodes=4-4 myjob Using a Resource Requirements String in an LSF CommandPage Glossary External network node Extensible firmware interfaceFairshare First come first servedIntegrated Lights Out Image serverInterconnect Internet addressLSF master host Network Information ServicesManagement Processor Master hostParallel application Root Administration SwitchResource manager role RoleGlossary-6 Symmetric multiprocessingIndex-1 IndexIndex-2 Gdb, 4-1 GNUIndex-3 ModulefileIndex-4 Resource manager, 7-1role Serial applications