[f-nsp] MLXe slow on 10G doing inter-vlan routing

[Mike Allen]

Turn off ICMP redirects. Common problem in one arm routing configs. IMO it
should be the default setting but it's not.

Mike
On May 24, 2016 2:00 PM, "George B" <georgeb at gmail.com> wrote:

> Do you notice the same behavior with both UDP and TCP streams?
>
>
>
> On Tue, May 24, 2016 at 1:42 PM, Steven Raymond <
> sraymond at acedatacenter.com> wrote:
>
>> MLXe, very lightly loaded, with only one card BR-MLX-10Gx8-X.  Simplest
>> of configuration, layer 2 link with VE interfaces.  During inter-vlan
>> routing where the ingress traffic and egress traffic are on the same port,
>> say e1/2, the input traffic level will be fine, but the output traffic is
>> effectively maxed out at about 50Mbps (on a 10GigE port).  Shift hosts
>> around so that the egress is a different port on the same card, bang
>> problem gone and full expected traffic rates achieved.
>>
>> Tried numerous code versions, from 05800a to 05900a to 06000.  No
>> improvement.
>>
>> Of course tried rebooting, new optics, new cabling.  No errors
>> incrementing on the interface.  This is the silliest dumb problem in a long
>> time.  Any ideas are appreciated.
>>
>> Thanks!
>>
>>
>>
>> _______________________________________________
>> foundry-nsp mailing list
>> foundry-nsp at puck.nether.net
>> http://puck.nether.net/mailman/listinfo/foundry-nsp
>>
>
>
> _______________________________________________
> foundry-nsp mailing list
> foundry-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/foundry-nsp
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://puck.nether.net/pipermail/foundry-nsp/attachments/20160524/adcb15de/attachment.html>