QOS - LLQ/CBWFQ - no-buffer drops

[csco10387876]

Hi all,

Can anyone give me a clear description of what the total drops and no-buffer drops are when I do a show policy-map xxx ?

I would like to be sure that the no-buffer drops counter is for packet that overflow the class queue and the total drops are for both no buffer drop and congestion management drops.

btw, what we have now is LLQ, no WRED and the drops are only from total not from nu buffer counter as here :

Class-map: class-default (match-any)

832119 packets, 115909611 bytes

5 minute offered rate 417000 bps, drop rate 161000 bps

Match: any

Queueing

Output Queue: Conversation 141

Bandwidth remaining 25 (%)Max Threshold 64 (packets)

(pkts matched/bytes matched) 823098/115254375

(depth/total drops/no-buffer drops) 38/336826/0

here we have 336826 drops but no wred so they should be no-buffer drops no ?

Thanks for your help.

[mheusinger · Accepted solution]

Hi,

there are two reasons, why a packet can be dropped in CBWFQ without WRED:

1) the queue limit for that class has been exceeded

2) there was no buffer that could be allocated to store the packet despite the fact that the queue limit has not been reached.

Total drops counts all dropped packets for a class (reasons 1 and 2) and "no-buffer drops" counts buffer allocation failures.

The latter is an indication that you need to increase the buffers (show buffers -> misses) in the router. This is serious, because your software queueing system could not service a packet, which should have been enqueued according to your qos policy.

So besides a proper qos policy also buffer tuning might be necessary. To give you a hint, if this is needed, the additional counter was introduced.

Some insight into all the counters of "show policy-map interface" is f.e. given in

http://www.cisco.com/en/US/products/sw/iosswrel/ps5187/products_command_reference_chapter09186a008017cf12.html#wp1153337

Search on the page for "no-buffer" to find the relevant text section.

Hope this helps! Please use the rating system.

Regards, Martin

[csco10387876]

Hi Martin,

Thanks for your prompt reply.

This confirm what I thought ;-)