EIC TOP

How to use batch job

  • You can submit batch job on eic system.
    new systemold system
    submitqsubbsub
    display job statusqstatbjobs
    cancelqdelbkill
    job scriptfile difinition#$#BSUB

Job Class Configuration

  • You have to select job class when you submit job.
    Job
    Class
    RUNTIMEMEMORY LIMITMAX MEMORY
    LIMIT
    CPU
    sockets(cores)
    NUMBER OF JOB
    ETC
    A2hours(cputime)8GB8GB1(1)4eic
    B100hours64GB64GB1(8)4(32)eich1~eich8
    C20hours256GB256GB4(32)3(96)eich1~eich8
    D80hours256GB256GB4(32)1(32)eich1~eich8
    E20hours512GB512GB8(64)2(128)eich1~eich8
    F80hours512GB512GB8(64)1(64)eich1~eich8
    G10hours2.2TB2.2TB36(432)2 (864)eicp1~eicp8
    H40hours2.2TB2.2TB36(432)1(432)eicp1~eicp8
  • 4 sockets(32 cores) is available on C and D class on eich. C is limited within 20 hours,D is limited within 100 hours. 8 sockets(64 cores) is available on E and F.
  • G,H is for MPI or Hybrid on eicp, you can use 432 cores(18 nodes / 36 sockets). Limit your job within 125GB memory per node on G,H class,or your job will fail.
  • See following instruction when you use B~H class, you can get detail information.

Running job

  • Make a job scriptfile and run qsub command.
  • Please use A or B class for a single job(non parallel job).
    • How to write scriptfile
      AB~FG、H
      #$ -jc A
      #$ -ac n=1
      #$ -ac P=1
      #$ -cwd
      program name
      #$ -jc Job Class
      #$ -ac n=(8 or 32,64)
      #$ -ac P=Cores(same as n)
      #$ -cwd
      Program name
      #$ -jc Job Class
      #$ -ac n=432(basically 432)
      #$ -ac P=24(Basically 24)
      #$ -ac T=OpenMP threads(1 when MPI job)
      #$ -cwd
      program name
      #$ -ac T is only for G and H.If you define it with other classes, your job will be rejected.
    • Sample scriptfile
      #!/bin/csh
      #$ -jc A
      #$ -ac n=1
      #$ -ac P=1
      #$ -cwd
      a.out
  • Submit
    run "qsub scritptfile" when you submit.<(redicection) is unncessary.
     eic:~> qsub sample.csh
     Your job 307 ("sample.csh") has been submitted
    you can see status by qstat.
     eic:~> qstat
     job-ID     prior  name user   state submit/start           jclass  slots 
     --------------------------------------------------------------------------------
     307 0.55500 sample.csh testuser r     02/12/2015 17:48:10    A.default 32
  • Qsub Option(Bold is MUST)
    #$ -ac n=432Total cores(All job classes)
    #$ -jc GJob class(All job classes)
    #$ -cwd Run current directory(All job classes)
    #$ -ac P=24Cores per node(All job classes)
    #$ -ac T=1OpenMP threads(only G~H classes)
    #$ -o filenamedefine outout file name
    #$ -m eSend E-mail when a job finish
    #$ -M xxx@xxx.jpE-mail address

Display your job status

  • you can display by qstat.
     eic:~> qstat
     job-ID     prior  name user   state submit/start           jclass  slots 
     --------------------------------------------------------------------------------
     307 0.55500 sample.csh testuser r     02/12/2015 17:48:10    A.default 32
  • Run qdel your job ID when you cancel or stop your jobs.
     eic:~/UGE_TEST> qdel 307
     testuser has registered the job 307 for deletion
     eic:~/UGE_TEST> qstat
     eic:~/UGE_TEST>

Standard output

  • job's stdout is saved as "scriptfilename.oJOBID" like "sample.sh.o251".Then stderror is saved as "scriptfilename.eJOBID".

トップ   編集 凍結 差分 バックアップ 添付 複製 名前変更 リロード   新規 一覧 単語検索 最終更新   ヘルプ