140 likes | 254 Views
Research Computing with Newton. Gerald Ragghianti Newton HPC workshop Sept. 3 , 2010. What is the Newton Program?. Research computing support Infrastructure management Consultation Training Research Objectives Effectiveness Efficiency Capability. The Newton cluster.
E N D
Research Computing with Newton Gerald Ragghianti Newton HPC workshop Sept. 3, 2010
What is the Newton Program? Research computing support Infrastructure management Consultation Training Research Objectives Effectiveness Efficiency Capability
The Newton cluster “Normal” Linux compute cluster 295 computers 2500 processors 5TB RAM 40 Gbit/sec Infiniband 80 TB Storage Head node Lustre storage Storage server Lustre storage Storage server Lustre storage External network Ethernet network Compute node Compute node Compute node Compute node Interactive node Compute node Compute node Interactive node Compute node Compute node Infiniband network
Getting started SSH to login.newton.utk.edu using NetID Transfer files with scp, sftp, or FileZilla Display graphics with X11, xorg, or Xming Requires X11 “tunneling” through SSH client $ ssh gragghia@login.newton.utk.edu Password: *************** [gragghia@newton1 ~]$ ls Test.sge filename.txt • [gragghia@newton1 ~]$ w • 10:36:49 up 32 days, 15:07, 20 users, load average: 1.98, 1.81, 1.88 • USER TTY FROM LOGIN@ IDLE JCPU PCPU WHAT • gragghia pts/0 poltth Tue05 1:05 1.39s 1.39s-bash • mkzadd pts/1 bkg.engr.utk.edu Thu18 15:16m 0.06s 0.06s -bash • Krrrcccpts/2 ares.bio.utk.edu 03Aug10 3days 0.03s 0.03s -bash
Environment management Modules utility Manages environment variables and aliases User chooses applications and libraries to use Allows multiple versions to be available Example use: See available modules: “module avail” Load a module: “module add R” Unload a module: “module unload R”
Resource Management: The Grid Engine Accepts job requests Executable to run Execution time Parallelization RAM needed Finds available resources (compute nodes) Reserves and uses resources Returns output
A simple job Create a job request file. Submit job $ qsub job.sge Monitor job $ qstat -g t View result log files #$ -q short* #$ -cwd #$ -N Test uname –a sleep 30
More Sophistication: Array jobs »Run the same job multiple times Create data files (optional) $ ~gragghia/workshop/make_datafiles.sh Create a job request file with “-t” option: Submit job $ qsub job.sge Monitor job $ qstat -g t View result log files #$ -q short* #$ -cwd #$ -N Array #$ -t 1-10 md5sum data-$SGE_TASK_ID.dat
A parallel job: MPI Download the software: $ wget http://newton.utk.edu/workshop/hello.tar Extract the software: $ tar –vxfhello.tar Select MPI version: $ module add openmpi/1.4.2/intel Compile the application: $ cd hello $ make Create a batch submit file Submit the job #$ -N Hello #$ -q short* #$ -cwd -V #$ -peopenmpi* 16 mpirun hello sleep 30
Compiling and Installing Software Example: Fractal generator Find the software Transfer to Newton Direct: wget http://newton.utk.edu/workshop/gmandel.tgz Indirect: Download to workstation and scp (sftp) Extracting the source code Uncompressed: tar Compressed: gunzip or unzip Configure the software: $ ./configure –prefix=$HOME/gmandel Compile: $ make Install: $ make install $ wgethttp://newton.utk.edu/workshop/gmandel.tgz $ tar –vzxf gmandel.tgz $ ./configure –-prefix=$HOME/gmandel $ make install …
Commercial Applications Matlab Graphical (interactive) Batch mode (parallel): matlab –r <Function> SAS SPSS $ module load matlabt $ matlab $ matlab –r ‘TestFunction’
More Information Newton Program website: http://newton.utk.edu/ Program policies Documentation Meetings / support / consulting schedule Research Computing Mailing List: USG_HPCC@listserv.utk.edu Visit http://oit.utk.edu/workshops/eval/ Section ID: Newton_Cluster-5