Dual Xeon Clusters

Steffen Persvold sp at scali.com
Wed Apr 17 00:46:44 PDT 2002

On Tue, 16 Apr 2002, Abhishek sinha wrote:

> Hi list members,
> I am building a dual Xeon 4-node cluster; My understanding of
> hyperthreading leaves me to conclusion that it depends largely on the
> code to benefit from it. Otherwise in many cases the performance can
> become worse than before using a hyperthreaded Xeon processors. My
> question is ; Are there any benchmarks available for the benchmarking of
> the Xeon processors in hyperthreaded mode ? Will the normal benchmarks
> that we use ..work on these systems and would it give a fair glance at
> the power of the Xeon .? If not what other way can i find the
> performance of Xeon processors in a clustered env.
> I am using 2.2 Ghz Xeon processors on an E7500 chipset.

First of all you will have to use a 2.4.18 kernel with these E7500
motherboards. Second, if you take a look at the linux-kernel malinglist
you will find a patch (originally developed bu Ingo Molnar, enhanced a
bit by me) that will do some IRQ balancing on Xeon chipsets (with the
stock 2.4.18 kernel i860 and E7500 chipsets are only able to handle
interrupts with CPU0). I don't know if this patch has made it to the
2.4.19-pre kernels yet, but you can check them out too.

Finally, I have some bad news about HT. I haven't been able to get it to
work stable enough with 2.4.18 (haven't tested 2.4.19-pre). The thing is
that in the beginning all works fine, but after a random amount of time
things start to slow down. Suddenly you find yourself having 'top' using
50% system time which is not normal. Turning off HT in the BIOS solves

As a side note I can tell you that the PCI architecture on this chipset is
_much_ better than on i860 and you can expect it to perform well with
high speed interconnects (Myrinet, SCI, GBE).

  Steffen Persvold   | Scalable Linux Systems |   Try out the world's best
 mailto:sp at scali.com |  http://www.scali.com  | performing MPI implementation:
Tel: (+47) 2262 8950 |   Olaf Helsets vei 6   |      - ScaMPI 1.13.8 -
Fax: (+47) 2262 8951 |   N0621 Oslo, NORWAY   | >320MBytes/s and <4uS latency

More information about the Beowulf mailing list