I have a QOS issue

Ahmed Zein · ‎12-29-2009

Hello everybody

I need help in QOS issue, we have 2 branches A and B, A connected via MPLS "Frame-relay" with 2 Mbps connection and B the same like A but with 256Kbps connection, both of the 2 branches connected via the MPLS cloud with many other branches, but only A and B configured with QOS.

And here is the configuration

1-Router A

Class match-any voice

Match protocol rtp

Match access-group 101

Policy-map voice

Class voice

Set dscp ef

Priority 1024

Class-default

Set dscp default

Policy-map voiceA

Class-default

Shape average 2048000

Service-policy voice

Int serial 0/3/0.1

Max-reserved bandwidth 90

Service-policy out voiceA

Access-list 101 permit ip 20.1.0.0 0.0.255.255 20.1.0.0 0.0.255.255

Note:all voice lans have this range

2-Router B

Class match-any voice

Match protocol rtp

Match access-group 101

Policy-map voice

Class voice

Set dscp ef

Priority 128

Class-default

Set dscp default

Policy-map voiceB

Class-default

Shape average 256000

Service-policy voice

Int serial 0/1.1

Max-reserved bandwidth 90

Service-policy out voiceB

Access-list 101 permit ip 20.1.0.0 0.0.255.255 20.1.0.0 0.0.255.255

the problem

when there is a traffic from A to B "DATA not VOICE" it saturate the INBOUND link of branch B, ok and no voice calls initiated yet.

BUT, when a call initiated voice from B to A is good "outbound of B"

And voice from A to B "inbound to B" so so bad, WHY??

The inbound to B only come from A ok, and data marked with dscp=0 and voice marked with dscp = ef, and the Service provider assure the QOS through its network

Note both A and B policy maps match the pakets and mark it, I checked it using sh policy-map int

Note during saturation of data I traced from A to B using voice subnet and delay = 50 ms in all hops except the last hop "from service provider to B" it equals 300 ms

Any one has any idea??

Thnx for your time

babatunde_sanda · ‎12-29-2009

Hello,

Your problem is with this command

Int serial 0/1.1
Max-reserved bandwidth 90

You are only using 10 percent of the interface bandwitch for you classified and marked traffic.

By default 25 percent of the interface is resevered. That leaves 75% for your policy map. But you went and reservered more. So your policy map is using the remaining 10% since you reserved 90% for what you didn't explicitly classify and mark. (That command is there for you to shoot yourself in the foot which seems like you succeeded in doing). Cisco recommends not reserving more than 25% of the interface bandwidth.

If you are just trying to give your voice precedence over any other traffic (which you should), you could revert that configuration to the default and simply use you configurations. It could be as simply as

Class match-any voice

Match protocol rtp-audio (if you don't specify both audio and video will be marked)
Match access-group 101

Policy-map voice

Class voice

Set dscp ef

Priority 1024

class class-default

fair-que

random-detect dscp-based

Int serial 0/3/0.1

Max-reserved bandwidth 25

Service-policy out voice

Access-list 101 permit ip 20.1.0.0 0.0.255.255 20.1.0.0 0.0.255.255 dscp ef (Use this or the second line)

access-list 101 permit ip any any dscp ef
access-list 101 permit udp any any range 16384 32767
access-list 101 permit ip any any precedence critical

cheers.

If this helps, pls rate.