Cisco Eases Deployment and Management of World's Largest InfiniBand HPC Clusters

Cisco Systems announced that it will showcase at Supercomputing 2005 the new DDR-capable additions to the Cisco InfiniBand-based Server Fabric Switch (SFS) portfolio: the 144-port Cisco SFS 7012 and the 288-port Cisco SFS 7024, as well as the high performance Cisco SFS Subnet Management Software -- a key component of Cisco's HPC management framework, Cisco Superflex InfiniBand flexible copper cables, support for open source drivers, and the ability to run InfiniBand over a wide area network. Together, these capabilities make it easier for customers to build larger and more manageable server clusters. Supporting the World's Largest Server Clusters Cisco InfiniBand Server Fabric Switches support some of the world's largest InfiniBand server clusters, which include: Sandia National Laboratories with 4500 servers, College of Computing at Georgia Tech, University of Sherbrooke, University of North Carolina at Chapel Hill, the National Center for Supercomputing Applications, and the University of Oklahoma. Additionally, Cisco Server Fabric Switches have been deployed in hundreds of server clusters, including both research and commercial environments in the US, Europe and Asia-Pacific. Server Fabric Switch (SFS) Portfolio The Cisco SFS portfolio, which includes the new high density Cisco SFS 7012 and Cisco SFS 7024 switches, utilizes InfiniBand technology to provide a high performance unified fabric for connecting servers together into grids of compute resources. Coupled with the SFS advanced Ethernet and Fibre Channel gateway technology, this combination helps seamlessly connect these server grids with shared LAN and SAN resources connected through Cisco Catalyst switches or Cisco MDS 9000 storage networking switches. High Performance SFS Subnet Manager The new high performance Cisco SFS Subnet Management software is specifically designed for the most demanding high performance computing environments, where rapid cluster start up and ongoing diagnostic capabilities are critical. With the high performance Cisco Subnet Management Software, Cisco delivers: -- The only subnet manager proven to scale beyond 4000 server nodes -- Capability to bring a multi-thousand node InfiniBand cluster fabric up in less than one minute -- Database synchronization for cluster high availability In addition, Cisco has tested and certified the OpenIB driver stack for HPC environments, giving customers the option of using the Cisco commercial driver stack or the OpenIB stack. Cisco InfiniBand Server Fabric Switches can be seen in the following booths at the Supercomputing 2005 conference in Seattle, WA from November 14-17: Dell, IBM, Intel, Mellanox, Microsoft, Pathscale, Sun Microsystems, and TeamHPC. Cisco is also a technology contributor in a demonstration on the ability to carry InfiniBand traffic over a Wide Area Network. The demonstration uses Cisco ONS optical switches that help two server clusters to communicate over an 80 km fiber link.