SDSC - ENZO

Phil Andrews
San Diego Supercomputer Center

SDSC is planning to bring a 32-node IA64 cluster to the SC04 show booth and use it to mount a GPFS file system on disks in the StorCloud booth. We expect to use approximately 100 TB of disk and will provide the HBA's necessary to access FC disks across vendor switches. We will use this cluster to mount the GPFS file system across the TeraGrid network to SDSC and other Grid sites. We expect to have three 10 Gb/s links available from the booth and across the TeraGrid WAN. Prior to, and at the beginning of the show, the application ENZO will run on the 10-Teraflop system, DataStar, at SDSC. ENZO is heavily I/O intensive and writes approximately 1 TB/hour during multi-hour runs. If we can, we will write the data directly from DataStar to the GPFS file system at SC04 in Pittsburgh across the TeraGrid WAN. ENZO simulates galaxy formation and writes out large amounts of data for post-processing. On the show floor, we will run a visualization application to attempt as high a transfer rate from the disks as possible to the visualization application running on the cluster in the booth. Across the TeraGrid WAN, multiple sites will access the data from ENZO on the show floor to run post-processing applications. This should provide a true Grid, data-intensive application. If possible, we will also attempt to link to the DEISA grid in Europe, and serve up data to their systems.