[Beowulf] Re: Beowulf Digest, Vol 9, Issue 12

Håkon Bugge Hakon.Bugge at scali.com
Mon Nov 15 07:50:57 PST 2004


The following systems are _not_ Myrinet systems:


neither were their predecessors. This is exactly 100% of the Norwegian 
contribution on this years June list ;-)

At 12:33 AM 11/13/04, Patrick Geoffray wrote:
>Håkon Bugge wrote:
>>I know of systems on previous lists which are listed as Myrinet systems, 
>>but are actually only running Gbe. This might stem from the fact that 
>>some vendors would like to get as many system listed as possible. When 
>>they have benchmarked and submitted an N-node cluster with Myrinet, it is 
>>very tempting to submit another N-node cluster using the same node hardware;-).
>Can you see such systems on the last list ?
>The Top500 team asked us before the official release to check the Myrinet 
>entries and we can confirm that we shipped hardware for all of them.
>We have found a few mistakes in the past (Ethernet cluster with the same 
>Rmax than similar Myrinet cluster for example). Sometimes, results are 
>abnormal (Myrinet clusters benchmarked with Ethernet or very bad 
>efficiency because of bad HPL parameters), sometimes people are too busy 
>to even run Linpack and their machines are not on the list, but vendors 
>tricks are usually easy to spot.
>Patrick Geoffray
>Myricom, Inc.

Håkon Bugge
dir. +47 22 62 89 72
mob. +47 92 48 45 14
fax. +47 22 62 89 51
Hakon.Bugge at scali.com

Scali - www.scali.com
High Performance Clustering

More information about the Beowulf mailing list