|
 |

| Cluster |
Overview
The Beowulf Linux cluster is a 24 node distributed memory multi-processor system. It contains a head node and 23 computation nodes, each containing two 2.8 GHz Intel Xeon processors with 533 MHz system bus, 512KB of cache (per cpu) and 2 GB of RAM (per node). The nodes are interconected with Gigabit Ethernet. A public network connects the head node with the internet. The cluster uses Red Hat Linux version 8.0 as its operating system. It uses the Portable Batch System (PBS) and the Maui Scheduler to distribute the computational workload across the nodes. The system is optimized for running parallel computations using MPI (message passing interface), but can also be used for standard serial programs.
|
Installed Software
System software: Intel C++ compiler 7.0, Intel Fortran Compiler 7.0, GNU Compiler Collection (GCC 3.2-7), Java (J2SDK 1.4.1_02), Perl 5.8.0, Bioperl 1.4, R-2.0, modeller-7v7-1, MPICH 1.2.5, Sift 2.1, Joy-5.10, ATLAS 3.4.1, GSL (GNU Scientific Library) 1.6-1, Standalone BLAST, MySQL-connector-java 2.0.14, MySQL-connector-C++ 1.7.9_3, QTL Cartographer V.1.16,
Database System: MySQL 3.23.55
|
Running Jobs
| Jobs are submitted from the head node by creating a PBS job command file which specifies certain attributes of the job such as how long the job is expected to run and how many nodes of the cluster are needed (e.g. for parallel programs). MAUI then schedules when the job is to start running on the cluster (based in part on those attributes), runs and monitors the job at the scheduled time, and returns any ouput to the user once the job completes.
PBS job submission tutorial
Login to your account by ssh; Remote file transfer should also be done by ssh. An SSH client software is available at http://www.utmem.edu/ssh
Compile your code, create the wrapper PBS Batch script as in the example below;
Submit the PBS batch script by %qsub scriptname
Note: You are in the headnode by default after login, where you are only supposed to do Source Code Editing, Compilation and Job Submission through PBS. We want the headnode to act just as the control console. Lengthy jobs in the headnode will be killed.
Example batch scripts: For serial jobs:
#PBS -N jobname
#PBS -M name@abc.com
#PBS -m ae
#PBS -q queueName
#PBS -l walltime=hh:mm:ss
cd $HOME/your_directory
./your_executable_file
For parallel jobs:
#PBS -N jobname
#PBS -M name@abc.com
#PBS -m ae
#PBS -q queueName
#PBS -l nodes=your_request_number_of_nodes:ppn=2
#PBS -l walltime=hh:mm:ss
MPIDIR=/usr/local/mpich/mpich-1.2.5_intel
PATH=$PATH:$MPIDIR/bin
cd $HOME/your_directory
NP=$(wc -l $PBS_NODEFILE | awk '{print $1}')
mpirun -np $NP -machinefile $PBS_NODEFILE your_executable_file
Available queues for your job: short: for jobs estimated to run 20 minutes, high priority
medium: for jobs estimated to run 24 hours, medium priority
long: for jobs estimated to run 72 hours, low priority
verylong: no time limit, lowest priority
Useful Commands:
Show the status of all the nodes in the cluster: %pbsnodes -a
Display a relevant listing of both active and idle jobs: %showq
Check the Status of a PBS Batch Job: %qstat
Cancel a PBS Batch Job: %qdel jobid
Check the detail of the job: %checkjob jobid
Display the earliest possible start and completion times for a specified job: %showstart jobid |
|
|
|
|
|
|
 |