[openfabrics-ewg] Test results from Silverstorm for OFED1.0

Betsy Zeller betsy at pathscale.com
Fri Jun 30 14:41:29 PDT 2006


Karun - These PathScale numbers are much lower than what I would expect
- for example, on RHEL4 U3, using Netperf, we've measured ~2.9 Gb/sec on
large packet IPoIB throughput over a pair of 2.4 GHz systems with our
PCI-e card installed. The other numbers seem similarly out of whack.

Is it possible for us to talk on the phone, to see if we can identify
the difference between the two test setups? I'll mention that we'll be
out of the office till July 5, for the Fourth of July holiday.

My number is 650-934-8088, and we are located in California.

Thanks, Betsy
 

On Fri, 2006-06-30 at 08:40 -0400, Sharma, Karun wrote:
> "MMS <sandia.gov>" made the following
>  annotations on 06/30/06, 06:43:25
> ------------------------------------------------------------------------------
>                      NOTICE                     
> A zero day exploit in Microsoft Excel has recently been discovered.
> If you know the sender of the Excel file(s), please contact that
> sender and verify that they intended to send you the file before
> opening it. If you do not know the sender, do not open this Excel
> file as it may be malicious.
> Please contact the CCHD (845-2243) if you have any questions.
>                      NOTICE                     
> ==============================================================================
> 
> Hi:
>  
> We have tested OFED1.0 with Pathscale HCA (x86_64) and Mellanox HCA
> (IA64).
>  
> Observations:
> 1. The throughput figures for both IPoIB and SDP are quite low with
> Pathscale HCAs as compared to Mellanox HCAs.
> 2. Iperf hangs while testing SDP. Sometimes I observed machine hangs
> also. I am debugging this issue.
>  
> Attached are the detailed result.
>  
> Thanks
> Karun
>  
>  
> _______________________________________________
> openfabrics-ewg mailing list
> openfabrics-ewg at openib.org
> http://openib.org/mailman/listinfo/openfabrics-ewg





More information about the ewg mailing list