Versions Compared

Old Version 3

changes.mady.by.user Ivan Mitrović

Saved on

compared with

New Version 4

changes.mady.by.user Ivan Mitrović

Saved on

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

  • a12.q
    • queue for all jobs with no limitations
    • servers HP ProLiant SL250s (12)
    • parallel jobs can spread exclusively within individual nodes because Infiniband network is not available to them 
  • a16.q
    • queue for all jobs with no limitations
    • servers HP ProLiant SL250s
  • gpu.X.q
    • queue for jobs that use graphical processors (GPUs) with a maximum duration of 7 days
    • servers Dell EMC PowerEdge C4140
  • p20.q
    • queue for parallel jobs with a maximum duration of 7 days
    • servers HP ProLiant SL230s Gen8
  • p28.q
    • queue for parallel jobs with a maximum duration of 7 days
    • servers Lenovo NeXtScale nx360 M5
  • p28-big.q
    • queue for parallel jobs that  
      • require number of cores which is the multiplier multiple of number 28 and
      • duration up to 7 days
    • servers Lenovo NeXtScale nx360 M5
  • p28-long.q
    • queue for parallel jobs with a maximum duration of 30 days
    • servers Lenovo NeXtScale nx360 M5
  • test.q
    • queue for the needs of application testing with a maximum duration of 24 hours
    • servers HP ProLiant SL250s
  • vsmp.q
    • queue for the use of ScaleMP vSMP nodes or jobs that require a lot of RAM or processing cores on a single node
    • servers HP ProLiant SL230s Gen8.

...

Parallel environment is an SGE mechanism that enables multi-core processing for the needs of parallel jobs. Since SGE does not have the option of detailed core division on nodes, Isabella cluster has three parallel environments defined:  Paralelna okolina je mehanizam u SGE-u koji omogućava zauzimanje više procesorskih jezgri za potrebe izvođenja paralelnih poslova. Kako SGE nema mogućnost detaljnog definiranja rasporeda jezgri po čvorovima, na Isabelli su definirana su tri oblika paralelnih okolina:

  • *mpi

    • SGE nastoji rasporediti procese na istom čvoruprvo će ispuniti sva slobodna mjesta na određenom čvoru prije nego prijeđe na sljedećiwill try to assign processes on the same node

    • when a node is at full capacity, processes will be assigned to the next node

  • *mpisingle

    • SGE će raspodijeliti sve procese na istom čvoru

      broj traženih jezgri ne smije premašiti broj dostupnih jezgri na čvoru

      will assign all processes on the same node

    • number of requested cores must not exceed the number of cores on a node

  • *mpifull

    • SGE će zauzeti cijele čvorove

    • broj jezgri mora biti višekratnik broja dostupnih jezgri na čvoru (ovisno o redu poslova)

    • will reserve the whole node

    • number of cores must be a multiple of the cores available on a node

Parallel environments available on Isabella are assigned to queuesParalelne okoline na Isabelli su podijeljene po redovima:

  • a12.q
    • a12-mpisingle – sve jezgre se dodjeljuju na jednom radnom čvorumpisingle – all cores are assigned on a single node
  • a16.q
    • a16-mpi – općeniti paralelni poslovigeneral parallel jobs
    • a16-mpisingle – sve jezgre se dodjeljuju na jednom radnom čvoruall assigned cores are on a single node
    • a16-mpifull - jezgre se dodijeljuju 16 po radnom čvoru (broj jezgri mora biti višekratnik od 16cores are assigned as a multiple of number 16 (whole nodes are reserved)
  • gpu.X.q
    • gpu – općeniti poslovi koji zahtijevaju grafičke procesore
    • gpusingle – svi grafički procesori se dodjeljuju na jednom radnom čvoru
    • general jobs that require graphical processors (GPUs)
    • gpusingle – all GPUs are assigned to a single node
    • gpufull - GPUs are assigned as a multiple of number 4 (whole nodes are reservedgpufull - grafički procesori se dodjeljuju 4 po radnom čvoru (broj grafičkih procesora mora biti višekratnik od 4)
  • p20.q
    • p20-mpi – općeniti paralelni poslovigeneral parallel jobs
    • p20-mpisingle – sve jezgre se dodjeljuju na jednom radnom čvoruall assigned cores are on a single node
    • p20-mpifull - jezgre se dodjeljuju 20 po radnom čvoru (broj jezgri mora biti višekratnik od 20cores are assigned as a multiple of number 20 (whole nodes are reserved)
  • p28.q
    • p28-mpi – općeniti paralelni poslovigeneral parallel jobs
    • p28-mpisingle – sve jezgre se dodjeljuju na jednom radnom čvoruall assigned cores are on a single node
    • p28-mpifull - jezgre se dodjeljuju 28 po radnom čvoru (broj jezgri mora biti višekratnik od 28cores are assigned as a multiple of number 28 (whole nodes are reserved)
  • p28-big.q
    • p28-mpifull - jezgre se dodjeljuju 28 po radnom čvoru (broj jezgri mora biti višekratnik od 28cores are assigned as a multiple of number 28 (whole nodes are reserved)
  • p28-long.q
    • p28-mpi-long – općeniti paralelni poslovigeneral parallel jobs
    • p28-mpisingle-long – sve jezgre se dodjeljuju na jednom radnom čvoruall assigned cores are on a single node
    • p28-mpifull-long - jezgre se dodjeljuju 28 po radnom čvoru (broj jezgri mora biti višekratnik od 28cores are assigned as a multiple of number 28 (whole nodes are reserved)
  • vsmp.q
    • vsmp – jezgre se dodjeljuju na cores are assigned to ScaleMP vSMP čvorunodes
  • test.q
    • test – pristup testnom čvoruaccess to test node

Jobs can be assigned to any queue by command Postoji mogućnost pokretanja posla u bilo kojem redu korištenjem opcije :

-pe *mpi X

ili za or for mpisingle posaojob:

-pe *mpisingle X