News Item Dated: 22 November 2002
From: CNet  (11/21/02); Shankland, Stephen

The InfiniBand high-speed networking standard, which never fulfilled its promise to replace PCI technology and revolutionize business computing, may get a second chance from the supercomputing sector, which has pledged support to build supercomputers and related products using the technology. Many InfiniBand companies are using this week's SC2002 conference to unveil products: Paceline Systems, which counts Sandia National Laboratories and the University of Washington among its clients, is collaborating with MPI Software Technology to build Message Passing Interface (MPI), an open-source software application that controls data sharing between disparate systems. Meanwhile, Dell Computers says InfiniBand clusters are currently undergoing laboratory testing to see if they could be a viable choice for high-performance computing. In addition, InfiniBand was used to interconnect 128 units into a supercomputer at Los Alamos National Laboratory. "The idea of having a high-performance, low-overhead interconnect that everyone can agree on is pretty appealing," notes Illuminata analyst Gordon Haff.

InfiniBand's advantages include higher data transfer speeds than Ethernet or Fibre Channel; InfiniBand 4x offers data transfer speeds of 10 gigabytes per second (Gbps), while Fibre Channel is standardized at 2 Gbps and mainstream Ethernet at 1 Gbps. Although InfiniBand is not cheap, it could be used in Beowulf clusters, clusters of interconnected Linux machines that can scale to provide high-performance computing. Such clusters are harder to program than traditional supercomputers but use inexpensive software such as Linux and GNU software.

Interested? For further reading, check out this link: http://news.com.com/2100-1001-966777.html