Experience with PBSPro on a 4 x 512p Altix SuperCluster
NASA Advanced Supercomputing Division (NAS)
e-mail: dtalcott@mail.arc.nasa.gov
Received:
Rec. 30 May 2006
DOI: 10.12921/cmst.2006.SI.01.61-64
OAI: oai:lib.psnc.pl:604
Abstract:
The Columbia project at NASA Ames includes twenty 512 processor Altix systems. Four of those systems are joined together into a Numalink based globally accessible non-coherent memory Altix SuperCluster. PBSPro 7.0 and 7.1 have been used to schedule and run work on this system. In this presentation, we will go over features of PBSPro especially applicable to such systems, some problems we have encountered and how they have been addressed, and remaining, unresolved issues.
Key words:
Altix, Columbia, Linux batch processing system, mpiexec, PBS, PBSPro
The Columbia project at NASA Ames includes twenty 512 processor Altix systems. Four of those systems are joined together into a Numalink based globally accessible non-coherent memory Altix SuperCluster. PBSPro 7.0 and 7.1 have been used to schedule and run work on this system. In this presentation, we will go over features of PBSPro especially applicable to such systems, some problems we have encountered and how they have been addressed, and remaining, unresolved issues.
Key words:
Altix, Columbia, Linux batch processing system, mpiexec, PBS, PBSPro
