Alpha trounces Sun in Oz supercomp rematch
Sun servers eclipsed in number-crunching project
Compaq technology has been selected for an Australian supercomputer project after an attempt to use Sun kit went horribly wrong.
Australian Partnership for Advanced Computing has announced its use of Compaq AlphaServer SC to build what is set to become the largest supercomputer available to Australian researchers and industry, after Sun kit failed an acceptance test.
As previously reported, in November a terse note on the APAC site in November said: "The initial configuration of the APAC peak computing system failed acceptance tests in September 2000, so the process of acquiring a system has been restarted. It's hoped there will be a system available by the second quarter of 2001."
The result of this reopened bidding is that APAC will now use AlphaServer SC-based system including more than 450 Alpha processors instead of a cluster of four Sun E10000 computer nodes.
APAC's defection from Sun is all the more embarrassing for Sun because the firm made such a fuss of winning the contract in the first place, and issued a press release on the initial contract win in August that it must now bitterly regret.
Sun's press release said it had "joined forces with the Australian Partnership for Advanced Computing (APAC) to install a powerful computing system for Australian researchers and industry professionals. The system will be installed in the APAC National Facility at the Australian National University, which is the host institution for APAC."
The cringeworthy paragraphs roll onward: "The three-year agreement will see an initial commissioning of a 200 Gigaflop system in September 2000, comprising a cluster of four E10000 compute nodes, and will progressively upgrade this to over one Teraflop by mid-2002. This means a five to ten fold increase in the capacity of the largest computer systems available for research and education in Australia."
Oh dear. Still, according to Apac's press release, the initial system will be operational in April 2001, and implementation will be complete in October 2001. So timescales for the project haven't been hit that badly by a decision to ditch suppliers. Maybe the British government, which seems to be perennially immersed in IT cockups, can learn something from the Australian experience? ®