Lecture 2TTH 03:30AM-04:45PM

Dr. Jianjun Huhttp://mleg.cse.sc.edu/edu/csc

e569/

CSCE569 Parallel Computing

University of South CarolinaDepartment of Computer Science and Engineering

Outline• Clusters and SMP systems at USC CEC• Ways of using High Performance

Systems• PBS Job Queuing System• How to write Job file• How to submit, delete, manage jobs

submitted to Linux Cluster• How to submit a large Number of Jobs

Systems: NICK Linux OS

Hardware• 76 Compute Nodes w/ dual 3.4 GHz XEON 2ML2,

4GB RAM1 Master Node w/ dual 3.2 GHz 2ML2, 4GB RAM

• Topspin Infiniband Interconnect• Storage: 1 terabyte network storage

Software• Rocks 4.3 CentOS Base, OpenMPI, OpenPBS/Torque• Absoft Compilers • Intel Compilers• Bio Roll that includes the following bio-informatics

packages: HMMER, NCBI BLAST, MpiBLAST, biopython, ClustalW, MrBayes, T_Coffee, Emboss, Phylip, fasta, Glimmer, and CPAN

• Intel Math Kernel Library• TURBOMOLE• VASP• STAR-CD

Systems: Optimus

Hardware• 64 Nodes: Dual CPU, 2.0 GHz

Dual-Core AMD Opterons, Totaling 256 Cores

• 8GB RAM • 1 Terabyte of Storage in Headnode• Gigabit Ethernet InterconnectSoftware• ROCKS 5.1• OpenMPI • OpenPBS Scheduler• GNU Compilers

Systems: ZIA

SGI Altix 4700 Shared-memory systemHardware 128 Itanium Cores @ 1.6 GHz/ 8MB Cache 256 GB RAM 8TB storage NUMAlink Interconnect FabricSoftware SUSE10 w/SGI PROPACK Intel C/C++ and Fortran Compilers VASP PBSPro scheduling software Message Passing Toolkit Intel Math Kernel Library GNU Scientific Library Boost library

Other SystemsNataku 8 Nodes: Dual CPU, 2.0 GHz Dual-Core AMD Opterons, Totaling

32 Cores 16 GB RAM in Headnode, 8GB RAM in compute nodes Chemical Engineering machine for Star-CD

Jaws2 8 Compute Nodes w/ dual XEON 2.6 GHz, 2GB RAM Remaining parts of original Jaws cluster, currently being rebuilt 1 Terabyte attached storage

Dr. Flora’s 12 CPU VASP Cluster

Dr. Heyden’s MAC Cluster for VASP

Distributed Multiprocessor Cluster

HD1HD2 HD3

Front End Node

NFS

How can we utilize large high performance machines like these to

speed up applications?

Question

Ways of using Linux ClustersApp. Type1:

data1

data2

data3

dataK

Regular program Collect results

Each data set is computed in a function independently as a job and can be run independently on one CPU

Ways of using Linux ClustersApp. Type2:

compute1

compute2

compute3

Compute 4

Communication between processes

result

Parallel processes can be executed on multiple CPUs and can be summarized together in the main process

data

Parallel program

PBS System for ClustersPBS is a workload management system

for Linux clustersIt supplies commands for

◦ job submittion◦ job monitoring (tracing)◦ job deletion

It consists of the following components:◦ Job server (pbs_server)

provides the basic batch services receiving/creating a batch job modifying the job protecting the job against system crashes running the job

PBPBS System for Clusterssing◦ Job Executor (pbs_mom)

receives a copy of the job from the job server sets the job into execution creates a new session as identical user returns the job's output to the user.

◦ Job Scheduler (pbs_sched) runs site's policy controlling which job is run

and where and when it is run PBS allows each site to create its own

Scheduler Currently Nick uses the Torque/Maui Scheduler

OpenPBS Batch Processing

Maui communicates ◦ with Moms: monitoring the state of a system's

resources◦ with Server: retrieving information about the

availability of jobs to execute

Steps needed to run your first production code

Suppose your application experiments are: $myprog data1 2 30

$myprog data2 3 12 ...

Steps to use PBS:1. Create a job script for each running experiment

containing the PBS options to request the needed resources (i.e. number of processors, wall-clock time, etc.) and user commands to prepare for execution of the

executable (i.e. cd to working directory, etc.).

2. Submit the job script file to PBS queue qsub prog1.sh

3. Monitor the job

First example: job1.sh jobfile#!/bin/bash

#PBS -N MyAppName

#PBS -l nodes=1

#PBS -l walltime=00:01:00

#PBS -e /home/dgtest/dgtest0200/test.err

#PBS -o /home/dgtest/dgtest0200/test.out

#PBS -V

Export PATH=$PATH:yourdir/bin;

myprog data1 2 30

Where is your output file located?Where is the screen output?

Jobfile for use on ZIA #!/bin/s

#PBS -N helloMPI #PBS -o hello.out#PBS -e hello.err#PBS -l select=1:ncpus=4#PBS -l place=free:shared

cd /home/<username>/test

mpirun -np 4 /home/<username>/test/hello

PBS Options #PBS -N myJob

◦ Assigns a job name. The default is the name of PBS job script.

#PBS -l nodes=4:ppn=2◦ The number of nodes and processors per node.

#PBS -l walltime=01:00:00◦ The maximum wall-clock time during which this job can run.

#PBS -o mypath/my.out◦ The path and file name for standard output.

#PBS -e mypath/my.err ◦ The path and file name for standard error.

#PBS -j oe◦ Join option that merges the standard error stream with the standard

output stream

PBS Options #PBS -k oe

◦ Define which output of the batch job to retain onthe execution host.

#PBS -W stagein=file_list◦ Copies the file onto the execution host before the job starts.

#PBS -W stageout=file_list◦ Copies the file from the execution host after the job completes.

#PBS -r n◦ Indicates that a job should not rerun if it fails.

#PBS –V#PBS –V◦ Exports all environment variables to the job.

Procedure Use command line

◦ Use editor to create an executable script: vi myExample.sh Use first example code

◦ Make myExample.sh executable: chmod +x myExample.sh

◦ Test your script ./ myExample.sh

Submit your script:◦ qsub myExample.sh◦ remember your job identifier

i.e. 96682

Monitor / Control a JobCheck wether your job runs

qstat qstat –a

◦ check status of jobs, queues, and the PBS server

qstat –f◦ get all the information about a job, i.e. resources requested,

resource limits, owner, source, destination, queue, etc.

qdel job.ID ◦ delete a job from the queue

qhold job.ID◦ hold a job if it is in the queue

qrls job.ID◦ release a job from hold

ExerciseProblem: Given 10000 html pages, count

the frequency of all words and report it as: keyword frequeny

Keyword1 frequency1...

Use PBS to submit 100 jobs

How to submit 100 jobsTypical ways:

1. read file list2. for each file, create a job file, and submit it to the PBS queueWrite a bash script, which submits a jobs for

different datasetsWrite a perl script to submit jobsWrite a C program to submit jobs

Quick psubPsub is a perl script that can wrap a

command line program into a job file and submit to the cluster queue

>psub jobname.sh “ prog.pl –i=1” this will create a job file “jobname.sh” and

submit to the server for running. No need to edit a job file anymore

Local Disk of Computing NodeNormally, the computing node of clusters

can directly read and write files on NFS storage space

If your program has intense write-read operation, reading and writing to NFS directory will cause high traffics

Solution:direct your output and input to local directories at computing nodes and after execution, copy the results file to NSF directory

/temp, /tmp /state/partition1

SummaryTypeI parallel computing applicationHow PBS works in Linux Cluster ComputersHow to submit jobs to Linux clusters

HomeworkProgramming Problem: Given a html page,

count the frequency of all words and report it as:

keyword frequenyKeyword1 frequency1...

Use PBS to submit 100 jobs to count frequency for 10000 html pages in next Lab session.

Learn how to compile C programs on LinuxLearn how to create PBS job fileLearn how to submit jobsLearn how to submit multiple jobsLearn how to compile and run MPI program on

NICK

Homework