From Documentation
Jump to: navigation, search
Target usage: see below
System information: see gulper system page in web portal
System status: see gulper status page
Real time system data: see Ganglia monitoring page
Full list of SHARCNET systems

This cluster was decommissioned in 2013.

General information is a Myrinet cluster with 27 nodes.

  • each node contains 2 cpus (Opteron @2.4GHz), providing 68 cores
  • each node contains 2 GB per node
  • configured to prioritize low latency parallel applications

Suitable use

Low latency parallel applications such as mpi jobs requiring up to 1GB/core, specify "-f opteron" with sqsub. For mpi or threaded jobs requiring upto 3GB/core memory and involving intensive local write activity of large size data sets to /tmp specify instead "-f xeon" with sqsub. Note that nodes 24-28 contain dual socket, quad core processors with hyper-threading enabled, yet appear as 16-way nodes even though their effective computing power is limited to 8 cores per node.