Beowulf with Gigabit Ethernet

alex at alex at
Fri Jan 11 09:40:16 PST 2002

I think that in general point to point performance information has a
limited value, whatever vendors might quote on their web-page. SCI might
be performing pretty well if it comes down to just latency and bandwidth
between two machines, but it is a ring topology. If you have
more machines on a ring they will share that same bandwidth. 

The price performance ratio is dependent on what application you are
running plus how many nodes you need: 3D SCI is more expensive in
comparison to Myrinet while 2D SCI is cheaper.


> Patrick Geoffray wrote:
> > Hi Øystein,
> >
> > Øystein Gran Larsen wrote:
> > > Check our ScaMPI datasheet at
> > > It uses ping-pong performance to illustrate the performance of SCI, Myrinet and Ethernets.
> >
> > Which Myrinet hardware are you using in your test ? It looks like old
> > 1.28 Gb/s interfaces not sold since quite a while (more than a year).
> > If it's the 2 Gb/s interfaces, the PCI is obviously the bottleneck
> > (the MPI curve caps normally at 240 MB/s with this equipment on a good
> > PCI 64/66).
> >
> > You may want to precise in your document the Myrinet model and the PCI
> > characteristics of the test machines, and eventually the fact that the
> > Myrinet equipment used is no longer available. This is required for a
> > fair interpretation.
> >
> > Regards
> >
> > ----------------------------------------------------------
> > |   Patrick Geoffray, Ph.D.      patrick at
> > |   Myricom, Inc.     
> > |   Cell:  865-389-8852          685 Emory Valley Rd (B)
> > |   Phone: 865-425-0978          Oak Ridge, TN 37830
> > ----------------------------------------------------------
> > _______________________________________________
> > Beowulf mailing list, Beowulf at
> > To change your subscription (digest mode or unsubscribe) visit
> Hi Patrick,
> I agree that the data sheet could be more detailed. As far as I have been able
> to uncover the Myrinet card in question was a PCI64B with 4MB SRAM and 2Gbit/s
> link speed. The performance measurements were performed outside Scali and made
> available to us.
> >From time to time we try to find performance numbers of this kind for different
> interconnects, but we have so far not been able to find any such (official)
> results for Myrinet products. If they were available it could be expected of us
> to use them when we try to compare products. In their absence we must rely on
> and trust helpful individuals that have access to Myrinet systems for the numbers.
> If you can direct us at an official resource with such numbers we would be very
> grateful.
> By the way, this type of performance numbers for the latest release of our
> software is not on our web site yet, but it's on its way. The numbers for previous
> releases can be found in the performance section on
> Sincerely,
> Øystein
> --
> Øystein Gran Larsen, Dr.Scient mailto:gran at Tel:+47 2262-8982
> ---------------------------------------------------------------------
>        MPI·SCI=HPC -- Scalable Linux Systems --
>    Subscribe to our mailing lists at

More information about the Beowulf mailing list