Following the close of SC07, the supercomputing conference in Reno, Nev., last week, Cisco Systems Inc. and other...
vendors announced new InfiniBand products that support high-performance computing (HPC) environments.
Ethernet has been the foundation of Cisco's networking technology offerings, but the company is working to strengthen its InfiniBand fare for the growing high-performance computing (HPC) market, said Pramod Srivatsa, product manager for Cisco's Server Virtualization Business Unit.
In addition, Cisco and Hewlett-Packard Co. have been collaborating on HPC market products, developing and testing interconnected blade systems across low-latency, high-bandwidth InfiniBand fabrics, including the new HP BladeSystem c3000.
Several vendors other than Cisco and HP have elbowed their way into the HPC market recently, such as Intel Corp., Microsoft Corp. and Sun Microsystems Inc. -- and rightly so. According to the World Wide High-Performance and Technical Computing Server Forecast released by Framingham, Mass.-based research firm IDC in April, the total high-performance computing market will expand by 9.1% each year in revenue to reach $15.5 billion by 2011, with cluster-unit shipments expanding by 12.5% annually to a total of 425,712 cluster units by 2011.
SFS 3504 is Cisco's latest version of a double-data-rate gateway that connects servers with storage area networks (SANs) and network-attached storage (NAS). The SFS 3504 has up to four gateway expansion slots and multi-terabit-capable backplane for scaling to hundreds of servers in a single chassis.
Cisco's SFS 3504 gateways are hot-pluggable expansion modules, enabling capacity expansion to both Fibre Channel and Ethernet storage devices. By sharing aggregated gateway ports in conjunction with the 20-gigabit-per-second (Gbps) InfiniBand connection, clustered servers gain access to more bandwidth than an individually connected server, Cisco reported.
"The performance is significantly improved compared to similar offerings, but not much more expensive," said Srivatsa. "The top-of-cluster design keeps in mind deployment of many clusters, so the box is beefier than some other offerings."
Users can deploy more than 1,000 servers with just one 3504 and receive the same I/O functionality expected from four previous-generation SFS products, Srivatsa said.
In addition, Cisco announced development and validation of Reliable Datagram Sockets (RDS), an acceleration protocol to further scale Oracle's Real Application Clusters (RAC) 11g using InfiniBand. The combination of Cisco InfiniBand server fabric switches and RDS interconnect helps users deploy to 16 industry-standard x86 Linux servers.
Mellanox introduced new NetworkDirect drivers for Windows HPC Server 2008 that takes advantage of InfiniBand's native Remote Direct Memory Access (RDMA) capabilities for clusters. The new drivers add scalability and performance for mainstream enterprise and high-performance-clustered applications, the company reported.
Based on OpenFabrics.org's open source community development, the InfiniBand drivers for Windows HPC Server 2008 are available now in beta from Mellanox's Web site and the OpenFabrics driver repository. The drivers will be available for production to coincide with the general availability version of Windows HPC Server 2008.
Voltaire announced that its InfiniBand products are available for the new HP Cluster Platform Workgroup System and introduced the Grid Director ISR 2004, a new 20 Gbps, 96-port InfiniBand switching platform for high-performance unified fabrics in the data center.
Let us know what you think about the story; email Bridget Botelho, News Writer.
Also, check out our news blog at serverspecs.blogs.techtarget.com.