|The Leading Source for Global News and Information Covering the Ecosystem of High Productivity Computing / November 12, 2007|
The perspective that computer maker Dell brings to its strategy on the HPC market is similar to the one it brings to the personal computer market: give customers what they want in a way that's simple, efficient, and just "works." Dell understands how to create technology for mass audiences, and they are trying to use that knowledge to build a mass market for HPC.
Many HPC companies aiming at markets other than the high end in HPC are focusing on engineering and manageability; some are focusing on usability and productivity for users as well. Dell is focusing on these areas, too, but it brings unique qualifications to its drive to make what Kevin Noreen calls "orderability" dead easy.
Dell feels that their experience allows them to deliver unique value to the customer that starts with ordering, and goes all the way through getting the machine into production with real users. Their focus on delivery and orderability has led the company to create an American Merge Center where HPC systems are pre-deployed, and all the software is pre-configured, before the system is shipped to the customer.
And, according to Noreen, this approach is resonating with customers. "Many of Dell's cluster customers are repeat customers. The University of Alabama at Birmingham just ordered its third or fourth cluster from Dell, and another major customer is using their delivery and acceptance experience with us as the model for how to get a massive system up and going."
In addition to innovating on its own, Dell is moving to adopt industry standard initiatives that make the cluster deployment process easier for customers. One example is the Intel Cluster Ready program, conceived by Intel as a way of assuring customers that systems bearing the designation would work with similarly-designated software stacks. Dell is also working with ISVs such as ANSYS to construct system configurations that are specifically designed to support customers solving real problems. "For example, we are working with ANSYS to build system configurations that support customers solving small, medium, and large problems using Fluent," says Noreen. "These configurations are described to customers in terms relevant to the problems they are trying to solve, rather than in abstract processor counts and memory sizes, so that they can be confident that the systems they're buying will address their specific needs."
Dell is going after both scientific and business users. In support of the first group the company is beginning to explore the addition of visualization to their compute offerings. Right now you can find white papers on their web site explaining how customers can transform a Dell cluster into a visualization engine to drive a tiled display, for example. According to Noreen, "we're looking at whether we can build a turnkey product out of this, and what the right way is to do something like that for our customers."
When I asked about where the company was on the various accelerator technologies being explored right now, Kevin answered that they are focused on GPGPUs deployed in workstations more so than in clusters right now, but that this could change in the future as customer demand increases and the technologies mature.
Dell is seeing the bulk of demand from its customers in the 64 to 128 server range, numbers which have been fairly consistent according to Noreen. The servers are dual processor quad core servers now, which means that the typical Dell customer is deploying machines in the 512 to 1024 core range. These systems are, of course, based on Xeon and Opteron processors. "Dell has had Itanium offerings at various times in the past," says Noreen, "but right now the customer demand just isn't there."
Noreen does say that he is seeing quite a bit of traction for Windows CCS 2003 on Dell clusters, along with quite a bit of interest in dual-boot CCS/Linux environments. According to him the demand is spread across all of Dell's market segments, and over all system sizes. And although the raw numbers for CCS shipments are still low, they are definitely growing.
John West is a freelance technology writer, a frequent contributor to HPCwire, and the publisher of insideHPC.com. He's been involved in supercomputing for the past 15 years.