[ofa-general] ipoib performanec at smaller mtu

Ben ben at lanl.gov
Mon Mar 9 17:19:23 PDT 2009


I have a similar setup in datagram, and don't see a big difference.   
Are these numbers expected?  I guess I would like to understand the  
limitations.  It seams like these numbers are a bit below what the  
hardware is capable of even with lro off.

-Ben

On Mar 9, 2009, at 6:09 PM, Boris Shpolyansky wrote:

> I think for small MTU sizes you will see better performance with IPoIB
> in datagram mode.
>
> Boris Shpolyansky
> Sr. Member of Technical Staff, Applications
>
> Mellanox Technologies Inc.
> 350 Oakmead Parkway, Suite 100
> Sunnyvale, CA 94085
> Tel.: (408) 916 0014
> Fax: (408) 585 0314
> Cell: (408) 834 9365
> www.mellanox.com
>
> -----Original Message-----
> From: general-bounces at lists.openfabrics.org
> [mailto:general-bounces at lists.openfabrics.org] On Behalf Of  
> Benjamin M.
> McClelland
> Sent: Monday, March 09, 2009 4:19 PM
> To: general at lists.openfabrics.org
> Subject: [ofa-general] ipoib performanec at smaller mtu
>
> I was curious what other people are seeing for 1500 mtu performance on
> ipoib.  Current tests are showing that I get about 150-175MB/s no  
> matter
> how many streams.  If I go up to higher mtu, then performance
> significantly increases.  Due to the storage network that this ib is
> routing to, we are currently limited to 1500 mtu.
>
> ofed 1.4.1, connected mode, dual opteron dual core 1.8GHz, Mellanox
> Technologies MT25418 [ConnectX IB DDR, PCIe 2.0 2.5GT/s] (rev a0) fw
> 2.6.000, voltaire sm on switch
>
> lro disabled, send_queue_size=256
>
> thanks,
> -Ben
> _______________________________________________
> general mailing list
> general at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>
> To unsubscribe, please visit
> http://openib.org/mailman/listinfo/openib-general




More information about the general mailing list