Penguin works out on Big Iron

Penguin works out on Big Iron


A typical Linux server consists of a single Intel-based CPU and costs less than $3,000. Linux also is being used in mainframes and supercomputers to reduce costs and eliminate dependence on a single manufacturer.

One such approach involves clustering a large number of inexpensive PCs or servers and running them in parallel. Called a Beowulf cluster, the first such unit'consisting of 16 Intel DX4 processors connected by 10-Mbps Ethernet'was built at the Center of Excellence in Space Data and Information Sciences at NASA's Goddard Space Flight Center in Greenbelt, Md., in 1994. The success of the project spurred more than 100 universities, laboratories and corporations to build their own clusters.

Many server manufacturers now offer Beowulf technology. The National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign is setting up a two-cluster computer, with each cluster capable of one trillion floating-point operations per second. The first cluster consists of 512 IBM eServer x330 nodes, each with dual 1-GHz Pentium III processors. The second cluster will use 320 of Intel's new 64-bit Itanium chips.

Linux clusters have some advantages over other supercomputing systems, including lower initial cost and lower upgrade costs.

But although Linux typically is the OS chosen to run each of the linked computers, the OS itself does not yet scale up to supporting the number of processors used in a Beowulf cluster. A common commercial solution is to use Myrinet connections and software from Myricom Inc. of Arcadia, Calif., to link the cluster.

IBM is taking a different approach to the problem. In addition to installing clusters for its clients, it has developed the means to run Linux sessions on mainframes. Last summer, the company's Design Center in Poughkeepsie, N.Y., demonstrated a Linux port to the company's S/390 mainframes.

Stay Connected

Sign up for our newsletter.

I agree to this site's Privacy Policy.