Help require ....QoS Issue

Question

policy-map test
class ce_management
police 8000 8000 16000 conform-action set-prec-transmit 6 exceed-action set-prec-transmit 2
bandwidth 8
random-detect
random-detect exponential-weighting-constant 6
random-detect precedence 6 39 62 10
class voice
police 212000 2000 2000
priority 214 1000
class application_data
police 918000 34000 34000 conform-action set-prec-transmit 6 exceed-action set-prec-transmit 2
bandwidth 1183
random-detect
random-detect exponential-weighting-constant 6
random-detect precedence 6 39 62 10
random-detect precedence 2 12 31 5
random-detect precedence 0 39 62 10
class class-default
police 8000 8000 8000 conform-action set-prec-transmit 5 exceed-action set-prec-transmit 5
bandwidth 131
random-detect
random-detect exponential-weighting-constant 6
random-detect precedence 5 12 31 5
Above config. is apply to T1 link and below is T1 int. config.
interface Serial1/3/0:0
no ip x.x.x.x x.x.x.x
encapsulation frame-relay IETF
load-interval 30
tx-ring-limit 1
tx-queue-limit 1
frame-relay lmi-type ansi
max-reserved-bandwidth 100
end
voice class looks for prec. 4 traffic and is feed with prec. 4 traffic.
data class looks for prec. 6 traffic and is feed with prec. 6 traffic
default-class is feed with prec. 5 traffic.
management looks for routing protocol traffic and rest is matched in default.
When voice class feed to police rate, application data class and default-class are fed with marked traffic to the config. bw under resp. class i.e. total traffic approached 1536K which is T1 interface BW.
Then random drops are observed under class-default and under application data class for traffic which is remarked to prec. 2
Question:- each class traffic is not exceeding the allocated bw. then why the random drops present in application data class and default-class?
If we assume that traffic on link is approaching 90% of interface bw and hence drops are observed then why drops are observed for application data class?

Answer 1

Hi,
how do you measure the "90% of interface bw"? The load interval of 5 minutes or even when tuned to 30 seconds is insufficient to detect short time bursts. With real traffic there will be bursts. So the interface may only be overloaded for a second or two. Still in this case queueing gets involved and according to your WRED or the policer config there may be dropped packets.
Can you provide the output from "show policy-map test interface Serial1/3/0:0"? This will allow us to see, why the packets are dropped.
In addition you reserve 100% of interface bandwidth for your classes. This will not work, because there are f.e. layer 2 keepalives, CDP, and some other router generated packets, which will NOT be placed in the classes. Cisco recommends 75% (factory default), you might increase this value, but I would not get above 90-95%
Also your markings are quite unusual. Are you f.e. aware, that routing updates from the router are marked with Precedence 6 and will end up in your data class?
Best practice would be (simplified):
Prec 6 - routing
Prec 5 - voip
Prec 4 - video
Prec 3 - voip signalling
Prec 2 - important data
Prec 1 - data class
Prec 0 - rest of traffic
Hope this helps! Please rate all posts.
Regards, Martin