very high bandwidth, low latency manner?

Joachim Worringen joachim at lfbs.RWTH-Aachen.DE
Thu Apr 18 05:29:34 PDT 2002

Markus Fischer wrote:
> On Thu, 18 Apr 2002, Joachim Worringen wrote:
> >Markus Fischer wrote:
> >If you could give some numbers, it would help very much. And which kind
> >of communication pattern is used in this application? Which MPI
> >communication calls, which message sizes?
> >
> >I am the author of SCI-MPICH. I do not understand the meaning of this
> >sentence of yours ("applies for the same statement"). What are you
> >refering to?
> >
> to have "the best performing"

Please indicate where you read this statement for SCI-MPICH. If at all,
it says "best performing of the evaluated implementations" in a certain
context. I do not (contrary to Scali) say that SCI-MPICH is the solution
to all your problems. Don't quote me wrong, please.

> >MPICH libraries, if you can not publish the source code.
> >
> >My bottom line is: I do not consider it good style to publically blaim a
> >product for bad performance without having checked back with the people
> >behind this product, and being a consultant for another product at the
> >same time.
> the first message was a follow up to other messages as a real
> world application performance issue. As I stated
> earlier, I did not focus on bringing this application to the max
> on every system, but to use an existing system and see how it goes.

This is perfectly understood. But if you experience strange results
which do not relate to what you would expect, wouldn't it be a good idea
to ask (in this case the Scali guys): "Hey, what is the reason for this
ugly numbers?" before publically announing "This technology is not able
to scale beyond 8 nodes for a simple application!". You won't deny that
there are numerous counter-examples.

> I also can read between the lines of some postings, too.

That doesn't help the reader of your postings.

BTW, it's enough to post to the mailing list, CC is not required.

 regards, Joachim

|  _  RWTH|  Joachim Worringen
|_|_`_    |  Lehrstuhl fuer Betriebssysteme, RWTH Aachen
  | |_)(_`|
    |_)._)|  fon: ++49-241-80.27609 fax: ++49-241-80.22339

More information about the Beowulf mailing list