This page introduces examples of job scripts for OCTOPUS.
Even if you do not know how to write a job script, it is possible to use OCTOPUS enough just by copying the example introduced here and slightly changing the value.

 

Scripts are categorized by purpose and by a node, so please select the ones that are likely to be helpful for yourself.
If you do not understand well, please read the following page first.

How to Describe a Job Script

 

Toggle Title
If the script does not run successfully"]

First of all, please check the following.

 

1.Do you insert "#!/bin/sh" in first line?
 
2.Do you insert the blank line at the end of script?
1 and 2 are script specifications and are required.
 
3.Is there a misspelling?
 
4.Is there extra space in the script?
For example, "#PBS -l cpunum_job=2, elapstim_req=0:10:00".
If you insert a space after the comma, an error will result.
 
5.Is the specification of resource amount correct?
Each system has an upper limit that can be specified according to the job class.
Please see the job class table for the detail of limit.

 

OCTOPUS job class table

 

6.If it still does not work

In that case please contact system @ cmc.osaka-u.ac.jp

 

 

1.one node execution

1-1.one node execution on a CPU node

Calculate 1 hour using 1 OCTOPUS:CPU node.
It is the basic job script to use OCTOPUS.
A reference page when writing this script: How to Describe a Job Script

 

1-1'.one node execution on a CPU node in "DBG" job class

Calculate 10 minutes using 1 OCTOPUS:CPU node.
DBG is a short-time job class for debugging purposes. The elapsed time can be specified up to 10 minutes at the longest, but the waiting time is reduced because the job rotates quickly.
A reference page when writing this script: How to Describe a Job Script

 

1-2: one node execution on a GPU node

Calculate 1 hour using 1 OCTOPUS:GPU node and 4 GPUs.
Jobs are assigned to GPU nodes by setting "gpunum_job" to 1 ~ 4 in the -l option (line 3).
A reference page when writing this script: How to Describe a Job Script

 

1-3: one node execution on a Xeon Phi node

Calculate 1 hour using 1 OCTOPUS:Xeon Phi node.
In the -q option (line 2), specify the queue "OCTPHI" for the Xeon Phi node.
A reference page when writing this script: How to Describe a Job Script

 

1-4: one node execution on a large-scale shared-memory node

Calculate 1 hour using 1 OCTOPUS:large-scale shared-memory node.
In the -q option (line 2), specify the queue "OCTMEM" for the large-scale shared-memory node.
A reference page when writing this script: How to Describe a Job Script

 
 
 

The following scripts are examples for the OCTOPUS:CPU nodes. When submitting a job to other nodes, refer to Examples 1-1 to 1-4 and rewrite the corresponding part.
 

2.in-node parallel execution (OpenMP / automatic parallelization)

Execute 24 cores parallel computations within one hour by automatic parallelization on 1 OCTOPUS:CPU node.
A reference page when writing this script:OpenMP/自動並列化利用方法

 

3.Use intelMPI (4 node × 24 Core parallel computing)

Execute 4 nodes × 24 cores (total 96) parallel computations within one hour by intelMPI on OCTOPUS:CPU node.
4 parallels between nodes and 24 parallels between cores are both parallelized by intelMPI
A reference page when writing this script:How to use intelMPI

 

4.Use intelMPI + OpenMP / automatic parallelization (4 node × 24 Core parallel computing)

Execute 4 nodes × 24 cores (total 96) parallel computations within one hour on OCTOPUS:CPU node.
4 parallels between nodes is parallelized by intelMPI, and 24 parallels between cores is parallelized by automatic parallelization.
A reference page when writing this script:How to use intelMPI+ノード内並列 利用方法

 

5.Use OpenMPI (4 node × 24 Core parallel computing)

Execute 4 nodes × 24 cores (total 96) parallel computations within one hour by OpenMPI on OCTOPUS:CPU node.
4 parallels between nodes and 24 parallels between cores are both parallelized by OpenMPI
A reference page when writing this script:OpenMPI利用方法