[c-nsp] 2960S drops/packet loss

[Mack McBride]

On the 2960 the buffer depth is 77 or 78.
On the 2960S the buffer depth is closer to 40 (don't have an exact figure).

The buffers on these are extremely small for a gig/10GE switch (less than 1ms).

LR Mack McBride
Network Architect

-----Original Message-----
From: cisco-nsp-bounces at puck.nether.net [mailto:cisco-nsp-bounces at puck.nether.net] On Behalf Of Anton Kapela
Sent: Sunday, December 25, 2011 1:25 PM
To: John Elliot
Cc: cisco-nsp
Subject: Re: [c-nsp] 2960S drops/packet loss

On Wed, Dec 21, 2011 at 9:41 PM, John Elliot <johnelliot67 at hotmail.com> wrote:
>
> Hi Guys,
>
> Have a pair of 2960's in a stack, one port(trunk) connects to another DC and we are seeing ~5% packet-lossand large output drops to this DC.

Wait up -- 2960, slight over-sub on trunk/uplink int, averages not close to line-rate, and yet tx discards? you don't say!

Depending on the situation, default queue configs (both with and without 'mls qos' globally enabled) are fairly naive -- I suggest:

-consider any need for actual QoS

-re-apportion queue allocations accordingly (as the defaults are often not ideal for 'real' networks with 'real' RTT's)

Here's what I've done on nearly every 2960G, S, 3650G, X, and 3750-whatever I've had to deal with given slight/moderate transmission oversub of any given int. The assumption in this config is that the network does not require fancy QoS, just high and less-high prio queues. That is, we can usually get away with a config that only includes enough queue isolation that bgp, ospf, voip/rtp all work stably, even when iperf and iscsi are slamming the shared links.

Consider the following global adjustments -- which, in a nutshell, take any input dscp values below 39 and place then in queue 1, 40 and up queue 2. It then sets discard thresholds for queue 1 and 2 to be 3200% and 3100% of max (usually ~100 packets are buffered), which is what provides a bit more burst ride-through. Previous versions of
2960/3560/3750 code didn't permit more than 100% of normal max buffering, but recent code (12.2(50)SE, later) seems to support permitting any one port having greater access to the anemic shared buffer pool.

mls qos map cos-dscp 0 8 16 24 32 46 48 56 mls qos srr-queue input bandwidth 9 1 mls qos srr-queue input threshold 1 90 100 mls qos srr-queue input threshold 2 90 95 mls qos srr-queue input buffers 95 5 mls qos srr-queue input priority-queue 2 bandwidth 5 mls qos srr-queue input cos-map queue 1 threshold 3  0 2 3 4 mls qos srr-queue input cos-map queue 2 threshold 2  5 mls qos srr-queue input cos-map queue 2 threshold 3  6 7 mls qos srr-queue input dscp-map queue 1 threshold 3  0 1 2 3 4 5 6 7 mls qos srr-queue input dscp-map queue 1 threshold 3  16 17 18 19 20 21 22 23 mls qos srr-queue input dscp-map queue 1 threshold 3  24 25 26 27 28 29 30 31 mls qos srr-queue input dscp-map queue 1 threshold 3  32 33 34 35 36 37 38 39 mls qos srr-queue input dscp-map queue 2 threshold 2  40 41 42 43 44 45 46 47 mls qos srr-queue input dscp-map queue 2 threshold 3  48 49 50 51 52 53 54 55 mls qos srr-queue input dscp-map queue 2 threshold 3  56 57 58 59 60 61 62 63 mls qos srr-queue output cos-map queue 1 threshold 3  5 6 7 mls qos srr-queue output cos-map queue 2 threshold 3  0 2 3 4 mls qos srr-queue output dscp-map queue 1 threshold 3  40 41 42 43 44 45 46 47 mls qos srr-queue output dscp-map queue 1 threshold 3  48 49 50 51 52 53 54 55 mls qos srr-queue output dscp-map queue 1 threshold 3  56 57 58 59 60 61 62 63 mls qos srr-queue output dscp-map queue 2 threshold 3  0 1 2 3 4 5 6 7 mls qos srr-queue output dscp-map queue 2 threshold 3  16 17 18 19 20 21 22 23 mls qos srr-queue output dscp-map queue 2 threshold 3  24 25 26 27 28 29 30 31 mls qos srr-queue output dscp-map queue 2 threshold 3  32 33 34 35 36 37 38 39 mls qos queue-set output 1 threshold 1 3200 3200 100 3200 mls qos queue-set output 1 threshold 2 3100 3100 100 3200 mls qos queue-set output 1 buffers 5 95 0 0 mls qos

Then the per-interface adjustment, which doesn't really matter until the link is nearing saturation:

 srr-queue bandwidth share 5 95 1 1
 srr-queue bandwidth shape  0  0  0  0
 priority-queue out
 mls qos trust dscp

Best of luck,

-Tk
_______________________________________________
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/