problems with 3com and intel 100MB cards

Dean Johnson dtj at uberh4x0r.org
Wed Oct 9 10:38:20 PDT 2002


On Wed, 2002-10-09 at 11:13, Marcin Kaczmarski wrote:
> 
> It is a proven fact that happened at some University in Germany that the
> newly bought super linux alpha dual cluster with 3com NIC ( I do not
> know the model of these cards in this case) simply failed to operate
> while trying to run very demanding scientifical calculations in material
> science just because of cards.

If it's Alpha cpu's, thats a whole different kettle of fish. Perhaps I
should have prefaced it with "With x86...". Running my molecular
dynamics benchmarks on our Alphas (Samsung 1100's) didn't show any
problems with the builtin interfaces, which I think were 3com, if I am
not mistaken.

> I am highly convinced that a
> server NIC which runs excellently in servers may be really absolutely
> not suitable for cluster that runs calculations, 

I am talking "server" in terms of server grade motherboard types and not
their usage. Server grade motherboards typically have much higher
quality, higher performance, and higher cost components. They often have
built-in maintainence functionality like the BMC. 

We have a whole bunch of consumer grade Dell dual cpu workstations that
have a variety of of 3com and Intel NIC's (eepro100 and e1000), plus
myrinet. We have given all the interfaces a serious beating with Amber
and NAMD, without a single NIC problem. Course, you mileage WILL vary.

> because you cannot
> compare the network load that you have on servers with the network load
> that appears while running in cluster, in case of cluster it is very
> very bigger. I`m sure of that. 

Each usage has different characteristics, even within the same domains.
My experience is in molecular dynamics and network usage is often the
dominant factor in scaling and overall performance. 

	-Dean




More information about the Beowulf mailing list