Interesting point.

But to add another wrinkle, the Gatekeeper is in the HQ site.  If you're 
supposed to mark all traffic, then the H323 RAS traffic between GK flowing to 
CCM wouldn't be marked.  

I'd never really thought about this.  

I'd be interested to know if Mark and Vik had any thoughts about this.  

Cliff

  ----- Original Message ----- 
  From: Ameha Negash Haile 
  To: [email protected] 
  Sent: Monday, April 13, 2009 9:45 PM
  Subject: [OSL | CCIE_Voice] QoS - H323 RAS signal marking


  Hello,

   

  First of all, thanks for the lots of great stuff on this list. I have been 
reading the various postings for very long. Now I have got some thought and I 
like to get your opinions on it.

   

  The question reads as “Mark SCCP, MGCP, H323 and SIP signaling traffic to 
CS3. Configure only on routers and CallManager.”

   

  The configuration on CallManager is straight forward. Regarding the 
configuration on routers, the solutions I have seen use access lists to capture 
interesting traffic (including h323 ras traffic at UDP port 1719), create 
policy-map to mark the traffic to CS3, and apply service policy in the inbound 
direction of voice VLAN interfaces on the various routers.

   

  As the above doesn't take into account traffic generated from the routers 
themselves, commands like 'ip qos dscp cs3 signaling' under VoIP dial-peers and 
'mgcp ip qos dscp cs3 signaling' on MGCP gateways are also used. I haven't seen 
any similar command for H323 RAS traffic generated from BR2 router and destined 
to the Gatekeeper on the HQ router as well as H323 RAS traffic generated from 
the gatekeeper (destined to the BR2 router or the CallManager gatekeeper 
controlled intercluster trunk).

   

  To provide a comprehensive solution that includes H323 RAS signaling traffic, 
what I think needs to be done is, combine the marking and queuing in the 
policy-map that is applied on the serial WAN interface in the outbound 
direction. While this takes care of the H323 RAS traffic between BR2 router 
(H323 gateway) and HQ router (H323 gatekeeper), to take care of the H323 RAS 
traffic from the gatekeeper to CallManager, we need to create a policy-map for 
marking and apply it on the server vlan interface in the outbound direction.

   

  A sample configuration for the portion from HQ to BR2 will be like:



  ! 

  Class-map match-any Voice

   match ip dscp ef

  class-map match-any Signal 

   match protocol h323            

   match protocol skinny

   match protocol mgcp

   match protocol sip 

  !

  ! Here we have accounted for all signaling traffic specified

  ! in the question. If we want, we may add 'match ip dscp

  ! cs3' and 'match ip dscp af31' before the ‘match protocol’ commands.

  ! The 'match protocol' commands will be used to capture leftovers.    

  ! 

  policy-map LLQ-Mark-HQ-BR2

   class Voice
    priority 96
     compress header ip rtp
   class Signal
    bandwidth 40

    set ip dscp cs3                  ! We do the marking here.
   class class-default
    fair-queue

  !

  policy-map MQC-768

   class class-default

    shape average 729600 7296 0

    service-policy LLQ-Mark-HQ-BR2

  ! 

  map-class frame-relay FRTS-768

   service-policy output MQC-768

   frame-relay fragment 960

  ! 

  interface Serial0/1.10 point-to-point

   bandwidth 768

   ip address 10.10.10.3 255.255.255.0

   frame-relay interface-dlci 310   

    class FRTS-768

  !

   

  By doing this, I believe we make sure all signaling traffic from HQ to BR2 
will be marked to CS3 before leaving the HQ router. We do the same on the BR2 
router also for traffic going to HQ.

   

  In my opinion, if we don't do the above, in addition to not properly marking 
H323 RAS traffic to CS3, we also miss H323 RAS traffic from the class-map 
Signal and the LLQ bandwidth guarantee won’t apply to it. I say this because 
what I have seen was that H323 RAS traffic is by default marked as 'DSCP 0', 
not even AF31. If this is the case, during congestion, RAS packets can be 
dropped big time!!

   

  I will appreciate your ideas on this.

   

  Thanks,

  Ameha Haile

   

  P.S. By the way, I find using 'match protocol' commands in class-map easier 
than using access groups and access lists. I don't have to worry about source 
ports and destination ports although knowing them is good. Could there be any 
other reasons, which I might be missing, to use one method over another for 
matching traffic?

   

   



------------------------------------------------------------------------------
  Create a cool, new character for your Windows Live™ Messenger. Check it out 

Reply via email to