<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<HEAD>
<META HTTP-EQUIV="Content-Type" CONTENT="text/html; charset=utf-8">
<META NAME="Generator" CONTENT="MS Exchange Server version 6.5.7638.1">
<TITLE>Re: [ofa-general] mlx4 second port lro issue</TITLE>
</HEAD>
<BODY>
<!-- Converted from text/plain format -->
<P><FONT SIZE=2>It should be. There is mlx4_en module param (num_lro) that controls it.<BR>
<BR>
<BR>
Boris Shpolyansky<BR>
Sr. Member of Technical Staff, Applications<BR>
<BR>
Mellanox Technologies Inc.<BR>
350 Oakmead Parkway, Suite 100<BR>
Sunnyvale, CA 94085<BR>
Tel.: (408) 916 0014<BR>
Fax: (408) 585 0314<BR>
Cell: (408) 834 9365<BR>
www.mellanox.com<BR>
<BR>
----- Original Message -----<BR>
From: Barry Mavin <Barry.Mavin@recital.com><BR>
To: Boris Shpolyansky; Shirley Ma <mashirle@us.ibm.com><BR>
Cc: Roland Dreier <rdreier@cisco.com>; general@lists.openfabrics.org <general@lists.openfabrics.org><BR>
Sent: Thu Sep 10 20:07:14 2009<BR>
Subject: Re: [ofa-general] mlx4 second port lro issue<BR>
<BR>
Is LRO on by default? If not how can we enable it?<BR>
<BR>
---<BR>
Regards<BR>
Barry Mavin<BR>
Recital Corporation<BR>
<BR>
<BR>
<BR>
> From: Boris Shpolyansky <boris@mellanox.com><BR>
> Date: Thu, 10 Sep 2009 19:31:35 -0700<BR>
> To: Shirley Ma <mashirle@us.ibm.com><BR>
> Cc: Roland Dreier <rdreier@cisco.com>, <general@lists.openfabrics.org><BR>
> Subject: RE: [ofa-general] mlx4 second port lro issue<BR>
><BR>
> Shirley,<BR>
><BR>
> Are you referring to actual Ethernet frame size or to TCP message size?<BR>
> If the port MTU set to 1500 it will reject Ethernet frames larger than<BR>
> this size, this has nothing to do with the LRO.<BR>
> LRO is a TCP offload that improves CPU utilization on the TCP receiver<BR>
> by combining multiple packets belonging to the same TCP stream to a<BR>
> single buffer and transferring it to the TCP stack as a single large<BR>
> packet.<BR>
><BR>
> Boris Shpolyansky<BR>
> Sr. Member of Technical Staff, Applications<BR>
> <BR>
> Mellanox Technologies Inc.<BR>
> 350 Oakmead Parkway, Suite 100<BR>
> Sunnyvale, CA 94085<BR>
> Tel.: (408) 916 0014<BR>
> Fax: (408) 585 0314<BR>
> Cell: (408) 834 9365<BR>
> www.mellanox.com<BR>
><BR>
> -----Original Message-----<BR>
> From: Shirley Ma [<A HREF="mailto:mashirle@us.ibm.com">mailto:mashirle@us.ibm.com</A>]<BR>
> Sent: Thursday, September 10, 2009 7:08 PM<BR>
> To: Boris Shpolyansky<BR>
> Cc: Roland Dreier; general@lists.openfabrics.org<BR>
> Subject: RE: [ofa-general] mlx4 second port lro issue<BR>
><BR>
> Hello Boris,<BR>
><BR>
> On Thu, 2009-09-10 at 14:16 -0700, Boris Shpolyansky wrote:<BR>
>> Dumb question: what was the MTU setting of the eth interface<BR>
>> associated<BR>
>> with port 2? Dropping jumbo frames has nothing to do with LRO - it is<BR>
>> plain layer 2 functionality.<BR>
><BR>
> ifconfig shows both port1 and port2 mtu are 1500. port 2 does't drop<BR>
> jumbo frames. The problem is the LRO is on for both interfaces so the<BR>
> interface will get large packet (packet size 1848). port1 can receive it<BR>
> and process it, but not port2. If I disables lro by reloading module<BR>
> with num_lro=0, then it will get small packet, and port2 works fine.<BR>
><BR>
> My question here is why port1 can work well for lro but not port2.<BR>
><BR>
> Thanks<BR>
> Shirley<BR>
><BR>
> _______________________________________________<BR>
> general mailing list<BR>
> general@lists.openfabrics.org<BR>
> <A HREF="http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general">http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general</A><BR>
><BR>
> To unsubscribe, please visit <A HREF="http://openib.org/mailman/listinfo/openib-general">http://openib.org/mailman/listinfo/openib-general</A><BR>
<BR>
</FONT>
</P>
</BODY>
</HTML>