80 likes | 235 Views
SURA Regional HPC Grid Proposal. Ed Seidel LSU With Barbara Kucera, Sara Graves, Henry Neeman, Otis Brown, others. Basic Plan. Strengthen SURAgrid to create the leading regional HPC environment Deploy numerous supercomputers across region Leverage regional investments in optical networks
E N D
SURA Regional HPC Grid Proposal Ed Seidel LSU With Barbara Kucera, Sara Graves, Henry Neeman, Otis Brown, others
Basic Plan • Strengthen SURAgrid to create the leading regional HPC environment • Deploy numerous supercomputers across region • Leverage regional investments in optical networks • SURA, NLR, RONs • 1 Gbit to many sites makes regional, national integration possible as never before • Coordinate deployment, operations • Major impact across region
Operational Plan • Tight integration of HPC systems • Globally shared file system • Common base software stack • Metascheduling • Machines respond both to local and regional needs • Majority of cycles locally controlled • Some fraction available for the regional use, coordinated training, preparation for codes to run at national centers
Primary Advantages • HPC Resource Sharing, load balancing • Regional (SURA sponsored), national training • Compatibility with national HPC centers • SURA underrepresented by 3.5-1 • Existing (NCSA, SDSC, NERSC, TACC, etc) • Future: LSU proposal, many others • Specific Projects • SCOOP, LEAD, Dynacode • Event-driven computing • Other projects much easier to develop with regional HPC support • IBM partnership
Software Deployment • Open Source • linux • Globus, Condor, Cactus, SAGA, MPICH, etc • Eclipse • Spruce, TeraGrid CTSS • IBM • AIX • GPFS-WAN • HPC Cluster software • ESSL
IBM Partnership • Hardware • Power5, Power6: very responsive • Software • Metascheduling, load balancing, migration of LPARS, MPI jobs • Development environment • Eclipse, Cactus, ESSL, Portals • Usage scenarios • Event-driven, DDDAS • Other HPC systems, software welcome and encouraged • TeraGrid model applies: all vendors connected
Financials • Very unusual value for major vendor • Price down to commodity levels • $1.2M system for $350K, including 3 years of maintenance (at roughly $112K) • SURA contribution likely if strong regional support is seen • Both hardware and personnel support possible • Some sites willing to help administer • LSU, others
Participating Groups • Expecting to participate: Kentucky, LSU, Oklahoma, UAH, Miami • Considering participation: TAMU, Houston, TACC, RENCI • Others