| gabriel rosenkoetter on Thu, 25 Apr 2002 13:55:21 -0400 |
|
On Thu, Apr 25, 2002 at 01:36:34PM -0400, Noah silva wrote: > Aren't the IPCs a bit slow for Paralell computing? Sort of, but if what you're doing is developing a purposely cross-architecture system, you take what you can get to test portability. > I thought about using banks of IPXs for this purpose, and while the IPXs > are cheap (I can get them about $20 a pop, or even free depending on the > time of year and way the wind is blowing), the memory is becoming more > expensive, and the AUI adaptors are either free or too expensive. Then > add in wanting a SCSI drive for each one, and the fact that I like to > upgrade the processors at $50 a pop, an old PII starts to get > cheaper. (and the 100baseT SBUS cards aren't cheap!) Yeah, but PIIs aren't good at the same things that sparcs are good at. (That is, different kinds of calculations.) One of the things I've been playing around with is placing processes based on what kind of math they do, how much memory they want, whether they're IO- or processor-intensive. (Placing them's easy, figuring out what kind of math they do is hard. And migrating them if you screw up is *really* hard... to the point of being not worth the effort if you need to migrate across architectures.) -- gabriel rosenkoetter gr@eclipsed.net Attachment:
pgp89IbquABDd.pgp
|
|