Guild Companies

While the server market is steadily losing air, one of the exploding markets for server capacity is clustered Linux servers based on various 32-bit and 64-bit Intel processors. While the Unix vendors who play in the high-performance technical computing market have been quick to pick up and peddle Linux clusters, which are relatively inexpensive compared to more parallel Unix servers on a gigaflops-to-gigaflops basis, to boost their sales, Linux clusters have not gone totally mainstream. Dell hopes to change that.

One of the best ways to get into a market, especially an exotic one like Linux server clustering, is to get academia, which is always an early adopter and proponent of new technologies, to get behind your product. Parallel clusters of Unix servers took off in academia in the late 1980s and became commercially viable in the mid-1990s as an alternative to very expensive vector supercomputers from makers such as Cray, Convex (eaten by Hewlett-Packard many years ago), NEC, Fujitsu, and Hitachi. Parallel machines have thousands of processors and complex, high-speed interconnection technologies that allow them to do some of the jobs that were formerly only possible to run on vector supercomputers. Linux clusters mimic the functions of these exotic parallel Unix machines--IBM's RS/6000 PowerParallel and HP's HP 9000 Hyperplex and AlphaServer Truclusters are perhaps the most popular machines installed in the world--but use cheap Intel server and interconnect components and free Linux software to create a somewhat less powerful parallel server than the top-end Unix parallel machines, but which can be one-tenth as expensive. Not surprisingly, many companies who have been ardent adopters of parallel supercomputers for the past decade are now looking at Linux clusters as a way to get more computing capacity for less money, much as in the last worldwide recession in the late 1980s research institutions and corporate IT organizations were looking around for a cheaper alternative to Cray and other vector machines and discovered parallel Unix servers.

Dell, which has sold a number of Linux clusters, wants a bigger piece of this action and it wants to get its name associated with high performance computing. To that end, the company is setting up the eponymous Dell Centers for Research Excellence at universities and, presumably, other influential centers of technical computing. The first center is going to be located at the University of Buffalo, a part of the State University of New York (SUNY) system of colleges and universities. SUNY Buffalo. That center will receive a 4,016-processor Linux cluster comprised of dual-processor PowerEdge servers running Red Hat's Linux and sporting over 16 TB of disk capacity. The cluster, which will be used for genomics research, consist of 1,900 two-way PowerEdge 1650 servers using Pentium III processors for number crunching and 100 two-way PowerEdge 2650 servers using Pentium 4 Xeon processors to be used as database servers. Middleware from Platform Computing, an established leader in Unix, Linux, and Windows clustering, will coordinate the parallel Fortran applications running on the cluster. The storage area network behind the cluster is being supplied by Dell's storage partner, EMC. The Linux cluster has a peak theoretical number-crunching capacity of around 5.6 teraflops (trillions of floating point operations per second). This will be the largest Linux cluster sold to date by any vendor--but it won't be long before IBM and HP beat this. Watch and see. The race is on to churn through that parallel Unix server installed base and sell these customers some new iron with Intel inside.