Last Friday provided an exception to this in the form of a most bizzare article entitled "Republic not even on the starting blocks in supercomputer race" by Kevin Cahill [subscription required].
The thesis of the article seems to be that without a supercomputer or two, Ireland cannot be taken seriously as a center of science and research.
I'm all for serious computing power, but the notion that the only way to get it is with liquid helium cooled supercomputers weighing in at 20-30 million Euros a pop is a specious argument that flies in the face of
both evidence and trend.
Mr Richard Hirsh of SFI (Science Foundation Ireland) is drawn over the coals for what seems to me to be an extremely sane advocacy of grid technologies.
Here is a quote from the article.
- "Mr Hirsh is an advocate of Grids, many computers locked together, but that is akin to saying that the state can get by with hiring a single combine harvester from abroad."
What?
I think Mr Cahill may need to revisit his understanding of grid technologies. Given that he has used a combine harvester analogy, I suggest he might like to start with von Neumann's curse.
He might also like this or this as a sign of what massive computational power will increasingly look like in the future. Typically, I might add, at a fraction of the 20-30 million Euro price point of the classic supercomputer design.
On his travels he may well spend some time in Google. A system that has massive computational power requirements. How many supercomputers do they have? Zero.
There are tasks where a monolithic supercomputer wins but in my experience, most computationally intensive problems are amenable to a good old fashioned domain decomposition approach. In these applications grids scale better than supercomputers in every dimension : money, power, common sense. The whole shebang.
If I had the time to, I'd love to pursue the idea of an XGrid from the (sadly moribund) xpipe project. The worlds fastest XML processing machine, capable of processing terabytes of XML per second. Now *that* sounds like worthwhile research to me and grids are the only sane way to get there.