[c-nsp] VoIP without QoS

[Eric Kagan]

> 
> On Tue, May 22, 2007 at 01:48:48PM -0400, Lamar Owen wrote:
> > On Tuesday 22 May 2007, Nassess, George wrote:
> > > I am in the process of extending our distributed VoIP 
> call center to a
> > > partner company, and their networking staff are extremely 
> adamant that
> > > they do not wish to implement QoS on their remote LAN, 
> the DS3 link that
> > > the voice traffic will traverse, or the core LAN in our shared
> > > datacenter. 
> > 
> > I too would like to see a good discussion of this, as I'm 
> getting prepared to 
> > implement VoIP here, on an 8540MSR core with Catalyst 5500-series 
> > distribution-access switches (using RSM's and RSFC's in 
> each 5500-series to 
> > provide dual layer 3 uplinks into the core, collapsing access and 
> > distribution); Cat 5500's and 8500's don't implement all 
> the things VoIP is 
> > supposed to require, but I'd like to see both sides, too.
> 
> 	I've typically "cheated" in doing voip QoS by rate-limiting TCP
> traffic in one direction.  This keeps the TCP traffic from taking the
> entire link and results in a basic reservation of traffic.
> 
> 	here's an example: (you may need to modify this based 
> on platform)
> 
> interface Serial0/0
>  description T1 to somewhere
>  ip address 1.2.3.4 0.0.0.0
>  rate-limit input access-group 106 1280000 4470 8000 
> conform-action transmit exceed-action drop
> !
> access-list 106 permit tcp any any
> !
> 
> 	The result is ~256k of reserved bw on a t1, enough for ~2x88k
> g711ulaw streams.
> 
> 	simulating the tcp loss with the rate-limit causes tcp 
> to think the
> link is smaller, yet leaving headroom for the udp bits :)
> 
> 	works well for a home network, you may need to adjust 
> depending on
> other streaming media applications that are udp based 
> (perhaps they need to get
> matched in your access-list 106) and depending on what you do.
> 
> 	As long as you don't have any true congestion and output drops
> on your interfaces (i assume you graph these?) you should be 
> ok without the
> qos stuff.
> g711ulaw streams.
> 
> 	simulating the tcp loss with the rate-limit causes tcp 
> to think the
> link is smaller, yet leaving headroom for the udp bits :)
> 
> 	works well for a home network, you may need to adjust 
> depending on
> other streaming media applications that are udp based 
> (perhaps they need to get
> matched in your access-list 106) and depending on what you do.
> 
> 	As long as you don't have any true congestion and output drops
> on your interfaces (i assume you graph these?) you should be 
> ok without the
> qos stuff.
> 
> 	- jared


Who or why are people against using a policy map that allows "up to x" of
bandwidth for the specified (ACL, class-map) but is available for all other
traffic when there is no VOIP ?  This seems backwards / worse to me to
restrict the pipe and loose that much available bandwidth for other apps
when there is no VOIP ?

Eric