Hi David,

I asked a similar question in the e1000 devlopment mailing list and this was
the responce I got.

"As far as the RSS hash mechanism the current implementation doesn’t
generate a symmetry between the destination and source values.  To achieve
that you would need to reduce the complexity of the RSS hash key to a
repeating 16bit value.  That would allow for both directions to generate an
identical RSS hash.  Flow director would also require similar changes to the
key in order to generate symmetry.  Otherwise 2 rules are needed to track
both sides of a flow."

So it seems like you need to set up 2 rules to redirect both sides of a flow
to the same rx queue so you need to know some of the following details on
the flow - the vlan tag, source/destination IP address and
source/destination port - you can wildcard some of these fields - see the
82599 manual section 7.1.2.7 on flow director filters (
http://sourceforge.net/projects/e1000/files/82599%20Developer%20Manual/2.0/82599_Datasheet_rev_2_0.pdf/download).
But be aware that IP fragmented packets miss the flow director filters
because only possibly the head fragment contains the L4 source/destination
port so these packets are assigned to a RX queue based on the output of the
RSS hash (even if you wild card the source and destination port in your
filtering rules).

Regards
Jonathan

On 25 March 2011 09:47, Luca Deri <[email protected]> wrote:

> David
> you cannot do that as RSS is not bi-directional
>
> Luca
>
> On Mar 25, 2011, at 9:14 AM, David Kwan wrote:
>
> Hi all,
>
> In my understanding, RSS of intel 82599 is not bi-directional, i.e "...This
> means that A->B packets go to queue X, and B->A go to queue Y, where X !=
> Y." by Luca at a reply on a question about "Problem about the TNAPI
> multi-channel".
> I now have a intel 82599 card with Flow Director, does anybody know how to
> configure the flow director such that the packets from the same TCP flow
> goes into the same queue?
>
> Thanks,
> David.
> _______________________________________________
> Ntop-misc mailing list
> [email protected]
> http://listgateway.unipi.it/mailman/listinfo/ntop-misc
>
>
> ---
> If you can not measure it, you can not improve it - Lord Kelvin
>
>
> _______________________________________________
> Ntop-misc mailing list
> [email protected]
> http://listgateway.unipi.it/mailman/listinfo/ntop-misc
>
>
_______________________________________________
Ntop-misc mailing list
[email protected]
http://listgateway.unipi.it/mailman/listinfo/ntop-misc

Reply via email to