Cheap, good ethernet cards?
Robert G. Brown
rgb at phy.duke.edu
Thu Apr 26 05:29:55 PDT 2001
On Wed, 25 Apr 2001 JParker at coinstar.com wrote:
> G'Day !
> Your Beowulf project page mentions you used Kingston Technology products
> for your cluster. Brahma. A quick serch shows that thay are priced about
> right. Would you not recommend them on your new cluster ?
The ones we used were Digital 21140 genuine tulip based. For a while I
wasn't impressed with the reports I got on the tulip list on the 21143
Intel/tulip based KNE cards, but it seems that several people on the
list are now willing to endorse them so I may give them another try.
I'm going to see if the vendor I'm dealing with can order them as
possible replacements for the 3c905's. I also had somebody write in
that the Netgear FA311 will "work" with the linux drivers posted on the
netgear website, but I'd prefer to run stock kernels in the cluster if I
can so I'm probably going to avoid this option unless/until they get
into a stock kernel. I'm going to have to check out the supported
device list in RH 7.1's 2.4 kernel.
> Jim Parker
> Sailboat racing is not a matter of life and death .... It is far more
> important than that !!!
> "Robert G. Brown" <rgb at phy.duke.edu>
> Sent by: beowulf-admin at beowulf.org
> 04/25/01 03:22 PM
> To: Martin Siegert <siegert at sfu.ca>
> cc: <beowulf at beowulf.org>
> Subject: Re: Cheap, good ethernet cards?
> On Wed, 25 Apr 2001, Martin Siegert wrote:
> > In short:
> > - RTL8139's are horrible
> > - Tulip cards are ok
> > - 3c905's are good.
> > I'd rather downgrade the processor speed soemwhat and use 905s than
> > save the the few bucks on the NIC. With respect to NICs it seems that
> > get what you pay for ...
> > Martin
> This matches my own experiences, although I would have rated old-style
> tulips as good as well. I'd settle for a tulip-based card in the right
> price range at this point, as you've put your finger dead on my fixed
> budget choice -- cheaper card or downgrade from 1.33 GHz to 1.2 or
> accept a slower FSB. Or monkey around with diskless/diskful
> configurations and so forth, which would likely work fine but takes more
> of my time and work. The problem is that this is about a 10% clock
> improvement and my applications are mostly CPU bound -- I really want to
> avoid an RTL card, but the 3c905 won't win much for me relative to that
> extra clock on a 266 FSB motherboard.
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb at phy.duke.edu
More information about the Beowulf