menu

Job Script

To run your job on our compute server, job script file including job setup information is required.

Job Script

Job Script contains "scheduler option (qsub option) " part, and  "command" part.

  • qsub option part: The line head is written as "#PBS”.
  • command part: execute the command in assigned shell.

Job Script : For MPI program

#!/bin/csh                  ←  Preferred shell
#PBS -q SINGLE              ←  Queue type
#PBS -oe                   ← Log file setting for joining error output and standard output into single log file.
#PBS -l select=2:mpiprocs=6   ← Resource Setting ( This is the setting for 12 MPI processes ) 
#PBS -N my-job              ← Name of your Job

setenv OMP_NUM_THREADS 1       ← Threads number for OpenMP. DON't omit this line and set the parameter as 1 if the program require to run as pure MPI.
setenv KMP_AFFINITY disabled

cd ${PBS_O_WORKDIR}        

mpiexec_mpt -np 12 dplace -s1 ./a.out        

Sample Script

Sample script is available in the following directory on UV3000.
Please use the resources after coping to your work directory.

 

uv: /work/Samples

Job Resource

Following is the sample of job resource setting.

MPI Only

- 2 vNodes (12CPU Cores , 256GB Mem)

-l select=2:mpiprocs=6

- 16 vNodes, (96CPU Cores , 2TB Mem)

-l select=16:mpiprocs=6

OpenMP

- 1 vNode (6 CPU Cores , 128GB Mem)

-l ncpus=6

- 8 vNode (48 CPU Cores , 1TB Mem)

-l ncpus=48 << (NON-MPI job only) script can assign the resource size by only ncpus parameter.

Hybrid

- 32 vNodes (192 CPU Cores , 4TB Mem) , 1Nodes per 3Process and 3x2 threads.

-l select=32:ncpus=2:mpiprocs=3

vNode resource table

1 vnode includes ( cpu corex6+  memoryx128GB).

vNode

1

2

4

8

16

32

64

128

256

Core Number

6

12

24

48

96

192

384

768

1536

Memory

128GB

256GB

512GB

1TB

2TB

4TB

8TB

16TB

32TB

<メモリサイズ指定時の単位について>:
mem=256gbのように,mb,gbを単位として記載することもできますが,
1mb=1024kbで計算されるため,予定の容量より多くのリソースを取得することになります.
この結果,以下に示すvnodeの個数を想定より多く必要とすることになり,ジョブの実行待ちが長くなるなどの影響があります

Queue Size (Revised 2017.7.28)

Queue

vnode

Core

Memory

Walltime

Number of simultaneous jobs(per user)

Priority

TINY

1-4

1-24

~512GB

6 hours

-(1)

-

SINGLE

1-2

1-12

~256GB

7 days

32(16)

150

  SMALL

4-8

24-48

~1TB

7 days

16(6)

130

 MEDIUM

8-16

48-96

~2TB

  72 hours
  4(1)

90

16-32

96-192

~4TB

LARGE

32-64

192-384

~8TB

48 hours

2(1)

30

 XLARGE  64-128  384-768  ~16TB  48  hours  1(1)  30

APPLI

1-2

1-16

~256GB

3 weeks

16(6)

110

LONG-S

1-81-48 ~1TB

2 weeks

3(1)

110

LONG-M

8-32

48-192

~4TB

1 weeks

1(1)

90

 LONG-L 32-96  192-576  ~12TB  5 days  1(1)  30

※ Starving(96 hours) + backfill