[Beowulf] Re: TOE on Linux

[Joe Landman]

Greg Lindahl wrote:
> On Tue, May 20, 2008 at 03:10:26PM -0400, Joe Landman wrote:
> 
>> StarCD.  Not big transfers, it doesn't move GB to its nodes.
> 
> By big I mean 100s of kbytes. I thought that this class of CDF

Ahh.... got it.  Yeah, it does that.  Actually a few megs between nodes 
(halo/shadows)

> computation, which uses a lot of memory, also does quite big transfers
> along the interfaces between the nodes.

Yes.  Big is of course relative ... :)

> 
> I agree with you that real application benchmarks are a fine thing.
> However, this one only has 2 datapoints for 1 application at 1 data
> size+node count with a potentially bogus MPI implementation, so it's
> bit hard to interpret.

Rather hard to draw inferences (well apart from a binary yes/no)

> 
>> As Ammasso is out of business, this is sadly nothing we could really use 
>> these days.
> 
> Never fear, there are a half-dozen vendors selling 10 gbit TOE cards.
> And nary a benchmark to be found.

(virtual coffee spew)

  :)



-- 
Joseph Landman, Ph.D
Founder and CEO
Scalable Informatics LLC,
email: landman at scalableinformatics.com
web  : http://www.scalableinformatics.com
phone: +1 734 786 8423
fax  : +1 734 786 8452
cell : +1 734 612 4615