SDSC and the TeraGrid: Next Steps

Patricia Kovatch
Seminar

Since the TeraGrid (TG) entered production in 2004, several initiatives have capitalized on the geographically distributed resources including wide area parallel file systems and metascheduling. Through several demonstrations at SuperComputing (SC), both technologies have been deployed on the TeraGrid. SDSC hosts GPFS-WAN, a 0.750 PB parallel file system mounted in production at ANL and NCSA. To allow more sites to mount GPFS-WAN, a demonstration is planned at SC '07 to export GPFS-WAN via pNFS to the TG. In an additional demonstration, we also plan to demonstrate the GPFS-HPSS interface and store a billion files in HPSS. Several metascheduling technologies have been evaluated and deployed including co-scheduling, automatic resource management and on-demand capabilities. Additional policy and technology development will allow these capabilities to be deployed in production across the TeraGrid. This presentation will review the history and promise of these initiatives.