[Beowulf] Three notes from ISC 2006
patrick at myri.com
Wed Jun 28 12:45:14 PDT 2006
Mike Davis wrote:
> Maybe I'm just too dense to understand. But, you've basically labelled
> Greg's post as spam.
Yes, I did. Telling me about a new white paper and about something that
I cannot know but I really should does fit my definition of spam. It was
borderline, I recognized that, and in retrospect I should have sent a
private email to Greg. But I still believe it was not appropriate.
> You've called their metric nonsense. You've
No, I did not. I called the marketing fluff around it nonsense. I have
been doing HPC communications long enough to know about message rate. I
also know that saying that the message rate is the most important metric
does not make sense. I would love to be shown wrong, though.
> criticized the published number they used that came from your company's
> product. For what?
For what ? To assess the technical merit of the white paper ?
The number they used are either obsolete or just plain wrong. The method
they use do not prove anything about the message rate being the
important metric in all of the benchmarks. I explained why I believed
that message rate is not an important metric. I gave numbers that I know
are correct. I am not doing the claim here, the charge of the proof is
not on my side.
> You've provided no new data.No reference to new data. When asked for a
> head to head test, you defer because in Europe such "marketing" (I don't
> necessarily consider benchmarks marketing) is not allowed. And, you've
Didn't you read my long post ? I think I provided enough technical data
to back my argument. Was I asked for a head to head test ? I didn't see
that in Kevin's post, but I don't mind doing tests. I don't have a
Myri-10G cluster available publicly with MX-1.2, but I can run any test
someone send me (with consideration for my time). You will have to trust
me though, and you should never do that.
I don't defer, despite being French. I just stated that *I* would not do
comparative marketing (and a white paper is a marketing material)
because I don't believe a vendor can do a fair comparison. It's the
rational in the market place in Europe for this good reason. I let this
responsibility to the academic world. They have written many papers on
that subject, and most of them include MX-10G results. Look at the
upcoming EuroPVMMPI or Cluster conference for example. I would never
believe the comparative results from another vendor.
> used far more bandwidth than Greg's orignal post which you critcized as
Yes, I did. That's always the case in this situation. But please note
that it was, I hope, technically valid and maybe, just maybe, quality
More information about the Beowulf