Cost drive takes on a new life form
A money-saving computing method used in the search for alien civilisations outside our galaxy has been adopted at cost-conscious Hong Kong universities.
Baptist University has become the latest higher learning institution to adopt the 'cluster' method in a $2.5 million project.
The idea is to connect ordinary stand-alone computers and pool their computing resources to generate enough speed and power to rival multimillion-dollar brand-name supercomputers.
The 'cluster' method was first proposed in the early 1990s by engineers at the US National Aeronautics and Space Administration.
It achieved renown when an international group, SETI (Search for Extra-Terrestrial Intelligence), began asking personal computer users worldwide to lend their computing resources for use in analysing radio signals received from outer space in the quest for patterns that may indicate intelligent messages.
SETI now has 3.8 million participants worldwide.
Baptist University is hoping to buy enough computer processors to build a 128-node system. Each node contains one or more central processing units, which is the 'brain' of a computer. For an equivalent brand supercomputer, the cost could be close to $10 million.
'We could buy at least 64 nodes, but we are bargaining for 128,' Baptist University assistant computer officer Morris Law Mow-man said.
In the past two years, the University of Hong Kong, the Chinese and City universities, and the University of Science and Technology have all developed powerful cluster projects as substitutes for supercomputers.
The latest University of Hong Kong cluster has 33 nodes and has replaced its US$1 million (HK$7.79 million) IBM supercomputer bought in the mid-1990s. Cluster programmes usually run the free Unix operating programme, which makes it unnecessary to buy expensive proprietary systems.
But Mr Law admitted a serious drawback with clusters was that if they crashed, work carried out up to that point could be irretrievably lost. Supercomputers allow for full recovery.
