|
|
|
|
|
The ECF's IBM SP arrived on Jan 11th, 2000. It moved to the Earth & Engineering Sciences building in March of 2000.
If you're interested in working on the SP, please visit our accounts application page. It's usually helpful if you've worked in a massively parallel environment in the past.
If you've just received your account, please see our getting started page for important information on how to log in, compile, and run jobs. It will describe the general organization of the machine, and will point out the ways that our setup differs from other sites that you may be familiar with.
Our IBM RS/6000 SP consists of a single frame housing 9 Winterhawk-II thin nodes and one Winterhawk-II wide node. Each node contains 4 processors and 2 GB of RAM. The nodes are connected via IBM's high-speed low-latency SP Switch. Gigabit ethernet connects the frame to the college network, and a HIPPI interface will provide a high-speed link to the ECF Cray.
8 of the thin nodes are connected to an SSA disk cabinet. These nodes serve a high performance, highly redundant 210 GB shared GPFS filesystem (/sparch) that is accessible to all nodes in the SP. This filesystem is under hierarchical storage managagement.
The wide SP node handles the gigabit ethernet and HIPPI network connections, and is the TSM (formerly ADSM) server for the ECF. It controls four tape drives on our STK Powderhorn tape silo, and has a dedicated portion of the SSA cabinet for the TSM spool. TSM provides the entire facility's backup services, and provides the HSM capability for the /sparch filesystem on the SP.