Companies in an expanding range of industries are using fully tested and certified cluster solutions with Linux to replace expensive proprietary systems.

For many IT managers, "HPTC" might still conjure up images of the massive systems used by people in white coats at government and research institutions, but the rise of Intel-based clusters running Linux has made high-performance computing available to companies in a wider range of industries, including financial analysis, computer-aided design and engineering, electronic design automation, geosciences, demographic research, and many others.

The Aberdeen Group recently projected that within the next few years, Linux will emerge as the primary operating system for HPTC. It also estimates that Linux is a viable choice for 80 percent of mid-range HPTC applications, as major ISVs move their applications to Linux.

Dan Cox, manager of Linux Cluster Programs in HP's Industry Standard Server Group, says the surging popularity of Linux for HPTC is easy to understand. "People who are running big RISC UNIX machines are saying, 'I've got this proprietary RISC chip, which I pay a fortune for, and this proprietary UNIX, which I also pay a fortune for. And I pay a fortune for maintenance on it too. And, yes, the thing does run like a screaming Mimi, but with a clustered dual-processor Xeon system running Linux, I could get about 75 to 80 percent of the performance at a fraction of the cost.'"

Cox says such systems are making HPTC available to more customers than ever before. "Take 32 nodes of HP ProLiant DL360s and put them together with Myrinet—high-bandwidth, low-latency interconnect—and it's going to perform like a Top 500 supercomputer for less than $250,000. That's why the Linux cluster phenomenon is booming."

When it comes to clusters, most IT managers are looking for ease of installation, deployment, and management, as well as the ability to scale out. But according to Cox, "If you buy a typical 32-node OEM cluster, what you get delivered will be two or three pallets of parts. Perhaps for an additional charge the cluster hardware will be put together and even have some software loaded, but the customer is left to handle the integration and acceptance-testing of the cluster to ensure it meets their needs."

HP, by contrast, is attempting to deliver an all-in-one solution, says Cox. "Customers can take it, plug it into the wall, and go. They won't have to worry about setup, testing, or loading the software because it's all been tested and certified by HP." The solution, Cox adds, will cable together properly and will be supported by HP's worldwide field service organization, as well as its worldwide consulting and integration services teams. These organizations can provide on-site architecture and performance planning to help customers choose the right cluster and size, and can provide expert education and training for both the hardware and software.

HP offers customers a choice of certified management software, including Scyld Beowulf Professional Edition and Scali Universe XE, Scali Manage, ScaliMPI applications, as well as Rocks cluster management software from the San Diego Supercomputing Center. Developed on the ProLiant DL360, Rocks depends on ProLiant as its reference platform.