[c-nsp] 3725 packet loss

[Rodney Dunn]

On Thu, May 31, 2007 at 05:27:24PM -0400, Scott Dunn wrote:
> Hi
> 
> We are seeing 2-4% packet loss on a 60Mbps link with 3725 when traffic
> reaches about 35-40Mbps (based 30s load interval) (FYI the circuit has
> been tested multiple times and is clean). I suspect that the traffic
> is bursting to 60Mbps for short period and causing the loss.

Easy to verify. Configure a MQC policy with a policer with a conform
action to transmit and an exceed action to transmit.

Then watch for matches in the exceed line of 'sh policy-map interface <blah>'.
If you see them you are bursting. Also, you would be seeing drops in the
'sh int <blah>' on output drops if you are bursting, overrunning the tx
side line rate, and filling up the outputqueue.

> 
> The Cisco command interpreter seems to indicate that the buffers need
> tuning as I'm getting lots of failures vs hits for the larger buffers.
> 
> VeryBig buffers, 4520 bytes (total 10, permanent 10, peak 13 @ 1w6d):
>      10 in free list (0 min, 100 max allowed)
>      84 hits, 685 misses, 11 trims, 11 created
>      685 failures (0 no memory)
> Large buffers, 5024 bytes (total 1, permanent 0, peak 4 @ 1w4d):
>      1 in free list (0 min, 10 max allowed)
>      14 hits, 671 misses, 131 trims, 132 created
>      671 failures (0 no memory)
> Huge buffers, 18024 bytes (total 1, permanent 0, peak 3 @ 1w4d):
>      1 in free list (0 min, 4 max allowed)
>      11 hits, 660 misses, 128 trims, 129 created
>      660 failures (0 no memory)

No they don't. And you shouldn't be hitting the system buffers for
transit traffic anyway. Those buffers are only for process switched
traffic.

> 
> I don't have any experience playing with buffer sizes and was hoping
> for some insight before experimenting,
> 
> Thanks
> 
> Scott
> _______________________________________________
> 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/