[Beowulf] recommendations for cluster upgrades

[Rahul Nabar]

On Tue, May 12, 2009 at 2:38 PM, Steve Herborn <herborn at usna.edu> wrote:
> Well to me it isn't wishing.  I guess it is all going to depend on how BIG
> your buy is,
>but I usually like to have them benchmark my application on
> whatever hardware they are proposing if it is different then my current.  If

Thanks Steve. Yes, that's right. We are probably a small customer for
them. Around 40 servers. So I doubt they will benchmark my application
for me. Do they actually take the effort to compile your scientific
app on several architectures? Just curious. Many of these apps don't
exactly run "out of the box" and it takes some work to get them
running. Or do you use some "standard" easily compiled package as a
benchmark analog?

>
> One thing about the Top500, they are the Top500 of those submitted, not of
> all systems out there.  There are many reasons why somebody wouldn't want
> their system submitted -- for example people who names are three letter
> acronyms.

True; we miss out on all the juicy details from the classified folks.

-- 
Rahul