I have a 3945 router with an ethernet wan connection. I'm applying qos and shaping to the interface. The counters are showing a considerable number of outbound discards. I am wondering why that is as the link is only 50% utilised. What should i check?
how do you measure link utilisation ? Most tools or graphs are not granular enough to account for traffic bursts that cause output drops, so even if link utilization (which is based on averages) seems low, output drops might still occur.
Can you post the QoS policy you have applied, as well as the output of 'show interfaces x' where 'x' is the interface that has the output drops ?
@Georg Pauweni would have thought that the use of shaping rather than policing would preclude the possibility of discards. At least until the buffers/queues fill up. How would i check if the queues/buffers are filling up? My QoS policy is:
description 20 MB Link To Carrier
ip address 10.42.250.1 255.255.255.248
ip wccp 62 redirect in
service-policy output GWIP-SHAPE
shape average 19800000
priority percent 18
bandwidth percent 15
bandwidth percent 5
bandwidth percent 5
bandwidth percent 27
bandwidth percent 4
bandwidth percent 1
you should compare the following two show commands
show interface <WAN>
! look for output drops here
show policy-map interface <WAN>
! look for drops in each class including class class-default if you have a parent shaper policy.
If drops appear only on show interface you are only facing drops caused by traffic bursts if drops are in the other show command means the software queues used to delay packets when shaping are sometimes full.
Hope to help
I'm revamping my policy as follows
class-map match-all CM-REALTIME
match ip dscp ef cs3 cs5 af41 af42
class-map match-any CM-CRITICAL
match ip dscp af31 cs6 cs7 cs2 af21 af22
class-map match-any CM-BULK-DATA
match ip dscp af11 af12
priority percent 33
bandwidth remaining percent 80
bandwidth remaining percent 1
bandwidth remaining percent 9
queue-limit 1600 packets
Do i need to include "random-detect dscp-based" in my policy map class statements?
Whats the recommendation as to when to include "fair queue" in the class statements?
Should i tune the "queue-limit" for each class?
Do i need to include a scavenger class (cs1)?
An item of interest from the show policy-map interface X output is:
Class-map: class-default (match-any)
10423137 packets, 9966185196 bytes
5 minute offered rate 8668000 bps, drop rate 0000 bps
queue limit 4000 packets
(queue depth/total drops/no-buffer drops/flowdrops) 0/4487/0/4487
(pkts output/bytes output) 10418650/9960660664
bandwidth remaining 9%
Is there a command i can run which would give me more detail? eg; dscp tag of dropped traffic, source or dest IP, or protocol? This is a gigabit ethernet interface shaped down to 20mb/s.
the most important fact is the number of output drops in class class-default is low compared to the total number of output packets in same class.
ratio is : 4487 /(4487 + 10418650 ) = 4,3048460362748757883543121423042e-4
There is no way to get more detais on dropped packets, however you have a GE interface with a 20 Mbps parent shaper and the class class-default gets only remaining percent 9% of those 20 Mbps.
I think this is acceptable in your scenario.
It is important to compare the output drops on show interface with the drops showed in show policy-map interface if the two numbers match or are very close all drops are performed by the shaper, otherwise if the total output drops at interface level is much greater then those seen in the show policy-map interface the drops caused by bursts of traffic are dominant.
Hope to help