[Beowulf] Gigabit switch recommendations

Bruce Allen ballen at gravity.phys.uwm.edu
Tue Mar 28 20:01:03 PST 2006


I suggest that you do some careful netperf testing of the 2724 before you 
purchase it.  We recently tested some 2716 switches with jumbo frames 
(MTU=9000) and found the performance to be well below the advertised 'wire 
speed'.  The response from Dell was 'when we say it's wire speed, we don't 
really mean it.  This switch is not designed for your application'!

After discussions with some of their technical people, the conclusion was 
that the switch had enough buffer memory to run wire speed with MTU=1500 
but not with MTU=9000.

Our testing results are here: 
http://www.lsc-group.phys.uwm.edu/~parmor/Network-performance/netperf.html

Caveat Emptor.

Cheers,
 	Bruce

On Tue, 28 Mar 2006, Joshua Baker-LePain wrote:

> My cluster and fileservers are moving to a new home shortly, and I'm going to 
> take the opportunity to put them on a private network.  Given everything is 
> gigabit and there's a lot of NFS going on, I'd like to use jumbo frames.  I 
> currently have 33 systems, and I'm not likely to grow by any huge amount (he 
> says with uncharacteristic optimism).  So I'm mainly looking at getting a 
> couple of 24 port switches.
>
> Given past research, I was about to go with the Dell 5324, but then I came 
> upon the much cheaper 2724.  The main differences for my purposes seem minor 
> (9018 byte vs. 9000 byte jumbo frame support, 8 member ports per aggregrated 
> link vs. 4).  Is there any compelling reason to go with the 5324 over the 
> 2724?  Does anyone have another recommendation in the same general price 
> range?
>
> Thanks.
>
> -- 
> Joshua Baker-LePain
> Department of Biomedical Engineering
> Duke University
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit 
> http://www.beowulf.org/mailman/listinfo/beowulf
>



More information about the Beowulf mailing list