Become a fan of Slashdot on Facebook

 



Forgot your password?
typodupeerror

Slashdot videos: Now with more Slashdot!

  • View

  • Discuss

  • Share

We've improved Slashdot's video section; now you can view our video interviews, product close-ups and site visits with all the usual Slashdot options to comment, share, etc. No more walled garden! It's a work in progress -- we hope you'll check it out (Learn more about the recent updates).

×

Comment: Re:Not in arXiv? (Score 3, Informative) 457

by GregPfister (#33185434) Attached to: Claimed Proof That P != NP

Circulating it among colleagues for review is exactly what he was doing. See the author's personal web page: http://www.hpl.hp.com/personal/Vinay_Deolalikar/

He says there that "The preliminary version made it to the web without my knowledge. Please note that the final version of the paper is under preparation, and is to be posted here shortly (in about a week). Stay tuned."

Comment: Re:Distributed, but not too connected (Score 1) 204

by GregPfister (#891512) Attached to: Distributed Operating Systems?
Thanks for the mention of "In Search of Clusters." An update: The second edition has been out since early '98. The link provided goes to the 2ndEd anyway. And it always was trade paperback; publisher's choice, and it's also a better way to publish the cool cover illustration.

A few comments on the "distributed OS" issue:

I think a reason for their lack of success has been (as I mention in "In Search") that they have been targetted at the wrong paradigm -- distributed systems -- instead of the right one: clusters. Until recently, don't forget, nobody knew from nothing about clusters. A good "distributed OS" is actually a great way to solve one of the biggest cluster headaches, management.

However, this is true only if you're doing a relatively small cluster. But one of the reasons people like clusters is the utility of "new" applications that scale obscenely well, like web serving. Unfortunately (!?) they scale far better than anybody's "distributed OS," so attempting to fix cluster management with a traditional distributed OS has a bit of a problem in many cases of interest.

About intermediate cases: The ones mentioned (Transputer, CM, Monarch (gak! somebody remembers that? How about the BBN Butterfly?), etc.) are indeed not intermediate cases. NUMA, however, really is an intermediate case. To get a NUMA system to work well, you actually do have to use localization techniques that are partway between a traditional SMP (everything's equidistant) and no shared memory (everything's still equidistant -- infinity = infinity).

Parts that positively cannot be assembled in improper order will be.

Working...