[asterisk-users] Setting QOS settings in asterisk and/or CentOS?

[Rich Adamson]

Nick Hoffman wrote:
> On Sat September 23 2006 06:14, Bob Amen <amen at oreilly.com> wrote:
> <snip>
>> which sets the TOS bit on all IAX, SIP and RTP packets. Using iptables
>> means that we can set up our rules on the router without using ACLs. Our
>> Cisco Cookbook (http://www.oreilly.com/catalog/ciscockbk/) has a nice
>> section on QoS (Chapter 11) and an appendix on TOS, etc. The author
>> advises not to use ACLs when possible as they take more CPU in the
>> router to implement and on a heavily loaded router can cause packet
>> delays. So here's what our config looks like:
> <snip>
>> Cheers,
>> Bob
> 
> 
> Hi Bob. I'm new to TOS and DSCP, but after going over your and Rich 
> Adamson's responses to Steve BerkHolz's question, I read up about them.
> 
> With what you wrote above, does this mean that your Cisco router(s) deny, 
> allow, and route traffic based on TOS/DSCP flags, and you don't bother 
> with traditional ACL rules like below?:
>     access-list 123 permit udp 1.2.3.4 ...

ACL's in cisco hardware can be used for pattern matching in addition to 
the old permit, deny, etc, functions.

Here's a working example from a cisco 1750 with QoS:

class-map match-all voice-rtp
   match access-group 103
class-map match-all www-traffic
   match access-group 105
!
!
policy-map voice-policy
   class voice-rtp
     priority percent 40
   class www-traffic
    bandwidth percent 30
   class class-default
    fair-queue

access-list 103 permit ip any any dscp cs3
access-list 103 permit ip any any dscp ef
access-list 103 permit ip any any tos min-delay
access-list 103 permit ip any any tos 12
access-list 105 permit tcp any eq www any

In the above, any packet matching the access-list 103 gets treated as a 
"voice-rtp" class, and in the policy map, is acted upon as "priority" 
(which means low latency queue) and can use up to 40% of the interfaces 
bandwidth.

The "bandwidth 384" statement on the interface "is" used by QoS to 
determine how much is actually going to be used for voip.

interface Dialer0
  bandwidth 384
  ip address negotiated
  encapsulation ppp
  dialer pool 1
  dialer-group 1
  service-policy output voice-policy
  ppp pap sent-username xxxxx_dsl password 7 136775499987

That bandwidth statement should be the "actual" amount of bandwidth 
available and not the value that your dsl/broadband provider says they 
provide.

Once the policy map is implemented, one can review the operational 
statistics by doing something like this:
C1750#show policy-map interface dialer0
  Dialer0

   Service-policy output: voice-policy

     Class-map: voice-rtp (match-all)
       1441504 packets, 191386680 bytes
       5 minute offered rate 0 bps, drop rate 0 bps
       Match: access-group 103
       Weighted Fair Queueing
         Strict Priority
         Output Queue: Conversation 136
         Bandwidth 40 (%)
         Bandwidth 153 (kbps) Burst 3825 (Bytes)
         (pkts matched/bytes matched) 0/0
         (total drops/bytes drops) 0/0

     Class-map: www-traffic (match-all)
       484061 packets, 341420115 bytes
       5 minute offered rate 0 bps, drop rate 0 bps
       Match: access-group 105
       Weighted Fair Queueing
         Output Queue: Conversation 137
         Bandwidth 30 (%)

Also, by doing the following:
C1750#show access-list 103
Extended IP access list 103
     permit ip any any dscp cs3
     permit ip any any dscp ef (1680 matches)
     permit ip any any tos min-delay (808709 matches)
     permit ip any any tos 12 (1 match)

one can "see" which piece of an access list is being matched. One can 
also see that both TOS and DSCP definitions can be used within the same 
access list. Its kind of a handy way to ensure voip phones and asterisk 
are properly configure and thus properly treated from a QoS perspective.

It should also be noted the above router is running v 12.2(4)T7 code. 
Cisco has made several changes to the syntax and parameters implemented 
in each version in the last few years.  In the newer IOS versions (for 
both switches and routers), the syntax and parameters are becoming much 
more standardized across all product lines.

The OP was specifically asking about QoS on a cisco switch, and without 
researching exactly what was implemented in "his" switch, there really 
isn't any way to give him a QoS template that would be accurate. For 
example, if I posted something that worked in the 12.4 code, its highly 
likely not to be acceptable syntax for 12.1 or 12.2.

Whether one uses access lists to do pattern matching is mostly 
immaterial "except" on a heavily loaded router. In my case, the 
processor utilization looks like:
C1750#show proc
CPU utilization for five seconds: 1%/0%; one minute: 1%; five minutes: 1%

where lengthy access lists would have almost zero impact.

For those that have read this far, it should be noted the implementation 
is a 3-queue policy (one for rtp, one for www, and one as the default). 
If the traffic for the rtp queue is low (or none), the unused bandwidth 
is automatically made available to other lower priority queues. In other 
words, the allocation of bandwidth to the various queues only occurs 
when demand is greater then the bandwidth available. Also, QoS can only 
be applied to "outbound" interfaces; inbound QoS must be done at the isp.

Rich