Linux to drive supercomputing cluster

An Australian supercomputing consortium is poised to lease a Hewlett-Packard Itanium 2-based computing cluster to service a multi-year contract to assist car maker Holden in new vehicle design and development. Victorian Partnership for Advanced Computing (VPAC) chief executive Dr Bill Appelbe told ZDNet Australia VPAC had opted to lease an HP Itanium 2 cluster consisting of 171GHz Itanium 2 processors, 64GB RAM, eight rx2600 nodes and more than a terabyte of storage to service the contract. The deal also includes full hardware maintenance over three years. The supercomputing cluster is expected to run on a Linux operating system in the medium-to-long term, with an HP UX system possibly used in the short term while some applications not yet ported to Linux are transitioned. Appelbe dismissed recent warnings from the SCO Group -- which has asked users to suspend their Linux development -- as "not a consideration" and described the vendor as "running on the FUD -- fear, uncertainty and doubt -- syndrome". Appelbe said that the system was expected to kick off in the coming months after relevant paperwork had been finalised. HP's solution beat out bids from rival heavyweights SGI, Dell and IBM after Holden and VPAC executives rated each proposal on a matrix encompassing price performance, risk, manageability and adaptability. Up to 30 engineers employed by VPAC, Holden or sub-contractors are expected to use the cluster -- based at Holden's new development facility at Fisherman's Bend in western Melbourne -- to conduct production engineering work. VPAC will manage the cluster via a high-speed leased line to its offices. Choices, choices
Appelbe said a key criteria for selection was price-performance. "We wanted the best-priced deal we could get for computing capacity on engineering jobs," he said, with the supercomputer expected to run a range of packaged engineering solutions such as LS-Dyna, Abaqus, Nastran and Star-CD. He said the cluster had to run a 64-bit architecture to deliver the capacity required to service the deal. Thirty-two bit architecture was determined to be inadequate, while a combined 32-64 bit solution was too fragmented for VPAC's application. "If you go to a solution that involves dealing with very large engineering problems, you need 64-bit architecture," he explained. The computing cluster is expected to allow engineers to develop very fine "mesh" with as many as half a million nodes and six degrees of freedom. This mesh is needed for simulations of design-crucial areas such as airflow within engine cylinders during activity. With air temperature and pressure varying widely at each small mesh node during engine activity, the ability to examine properties associated with these changes from a three-dimensional perspective is a crucial design asset. Appelbe noted that "that difference in air pressure impacts on performance. Achieving as close as identical airflow into all cylinders is a huge optimisation". He said the deal continued an ongoing "revolution in virtual engineering and virtual manufacturing," whereby traditional "rule of thumb" approaches to design were being replaced by highly-refined tools. Appelbe said another consideration of the deal was scalability, with VPAC envisaging a significant upgrade program. "Within a year, if we haven't added more capacity, we're not doing our job". He added the aggressive deployment of Itanium 2 solutions by vendors was rapidly increasing VPAC's options for future expansion. VPAC and Holden also assessed risk, examining each proposed solution against the danger of:

• a particular, necessary software product not running on the platform, or not delivering optimal performance;
• the machine having inadequate capacity;
• the machine being difficult or expensive to administer.

VPAC also wanted to ensure it was using a familiar operating system to ensure its existing skills base could be used effectively. Linux is widespread throughout its existing environment. Appelbe added the decision to go for a leasing arrangement was made primarily on a financial basis, as the entire computing power of the HP solution would be used to service the Holden contract. However, he also noted that "in high performance computing, the anticipated lifetime [of a system] is three years," with organisations who maintain their solutions beyond that date risking the loss of their competitive edge. "Computing power is doubling each year for the same dollars," he said. "The curve has even picked up over the last couple of years". He claimed the deal gave VPAC the best research and development capability in high-performance computing terms in Australia and helped the organisation achieve commercial self-sufficiency within two years of starting operations. The timing of the organisation's achievement was ideal, Appelbe said, as there was presently "a renaissance in supercomputing, with relatively new disciplines such as bioinformatics requiring heavy supercomputing grunt to open up commercial opportunities.

---