[c-nsp] changing buffer size on 4900M - discards

[Tóth András]

Hi Holemans,

The Catalyst 4900M Series can be deployed for 10/100/1000 server
access with 1:1 uplink to downlink oversubscription, mix of
10/100/1000 and 10 GbE
servers or all 10GbE servers in the same rack. The output drops are
caused by the egress buffers filling before the interface is able to
forward the traffic which could be caused by bursty traffic, although
the average rate is lower than 2 Gbps.

I would recommend to investigate if there are short bursts coming from
the 10G link which can cause the drops.
Also, check if you have flow-control coming from the 2G channel, which
can also cause output drops. Try disabling ingress flow-control on the
2G channel members and see if that resolves the issue.


Additionally, if the above does not help, I would suggest to check the
Queue-limiting and Queue Memory sections in the QoS guide:
http://www.cisco.com/en/US/docs/switches/lan/catalyst4500/12.2/54sg/configuration/guide/qos.html#wp1437305
http://www.cisco.com/en/US/docs/switches/lan/catalyst4500/12.2/54sg/configuration/guide/qos.html#wp1522202

Increase the queue limit gradually with the 'queue-limit'
configuration command under the policy-map class section. You might
try changing it straight to 512. If no policy-maps are configured yet,
create a new one and use queue-limit in class-default, then apply the
policy-map to the egress interface.

Best regards,
Andras


On Wed, Jul 27, 2011 at 1:35 PM, Holemans Wim <wim.holemans at ua.ac.be> wrote:
> We are seeing discards on a newly installed 4900M, probably coming from the fact that most input to the C4900M is coming from routers connected to it on 10G lines and is going out on a 2G etherchannel, although the total load on the 2G channel is just about 250-300 Mb/s. The 2G connection goes to an IPS that will be replaced before the end of the year but until then I have to find a way around the discards.
> Based on the fact that the 4900M is normally mentioned as a switch with a good buffer capacity (compared to 37xx switches, see also threads of begin this week), I wonder if  there is a way to change buffer size on the gigabit interfaces so that there will be less discards ? Anyone has a reference to a good document on buffer tuning (on 4900M) ? I know the 'buffers' command exists but for the moment I'm still trying to find out what buffers I should change (and into which values) to get rid of these discards.
>
> Greetings,
>
> Wim Holemans
> Netwerkdienst Universiteit Antwerpen
> Network Services University of Antwerp
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>