The review concludes:
Overall, I was quite impressed with both the Network product and the Framework SDK. If you have a bunch of Windows boxes sitting on desktops or feel like ordering a pallet-load or two from some discount vendor, I can’t think of an easier way to set up cost-effective compute farms. I did find the Workbench application somewhat confusing at first; nontechnical users would likely squeal “Argh!” when confronted with their first job template, but persistence and a bit of delving in the docs will either get them up to speed or let you create a cheat-sheet for them. And the documentation is quite clear and reasonably complete; the Framework SDK material is especially good.
In my admittedly small-scale testing, I often saw near-linear speedups when submitting jobs to the Network. Your mileage may vary, especially if you’re squirting gigabytes of data back and forth.
However, as I seem to say in just about every review, this isn’t magic fairy dust. You may have noticed me waving my hands rapidly over the Worker class (“computation happens here”), but none of this infrastructure attacks the hard problem of parallelizing your algorithms in the first place. Rather, think of this as a way to make the management and deployment of the programs you do eventually develop as painless as possible—at least until we get those quantum computers.
Actual article at:
http://www.acumeninfo.com/eprints/6338digipede.html?clientId=erights
Press release by Digipede:
http://www.prweb.com/releases/2006/1/prweb330903.htm
No comments:
Post a Comment