Re: [Diffserv-interest] DiffServ over ATM

I agree with Jeff. The intent of diffserv is make it unnecessary to mess 
around with parallel ATM VCs and the like - diffserv shares a single pipe
among multiple traffic classes. If each diffserv queue has its own VC,
diffserv is probably a no-op. Also, it's no surprise that IP forwarding is 
not a bottleneck on such slow links.

Now you know why some of us think that running ATM over wireless is
really, really pointless.

  Brian

Jeff Norman wrote:
> 
> John,
>   I know others may disagree or agree with me, but IMHO, queuing on
> multiple queues based on IP QoS/CoS/DiffServ and then queuing on
> another set of multiple queues based on ATM service categories is
> somewhat redundant.
> 
> Typically it's one or the other, or if both do exist then the ATM queues
> should be applying back-pressure onto the IP queues such that packets
> can get differentiated and conditioned by the IP queues as well.
> 
> It looks like to me in your situation, the back-pressure isn't happening.
> 
> Best Regards,
> --Jeff
> 
> In message "[Diffserv-interest] DiffServ over ATM", John Tillberg writes:
> 
> >Hello DiffServers!
> >I'm currently doing some simulations (using Opnet) of IP over ATM where I
> >use WFQ to differentiate between two separate traffic classes on a network
> >with low bandwidth links (2 Mbps). It appears that the WFQ mechanism doesn't
> >have any significant impact on the performance of the network. No matter how
> >I tune the weights, or even if I overload some of the links, I get almost
> >zero delay for both of my queues. However I do get delay on the ATM layer
> >(and I am aware of that I have to map my IP QoS classes onto proper ATM QoS
> >classes). I presume this is because the forwarding speed on the IP layer is
> >much higher than the bandwidth of the network. Hence every (conforming)
> >incoming packet will receive service on the IP layer (i.e. sent down to the
> >lower layers) right away even if there is a state of congestion on the ATM
> >layer. Have I misunderstood the matter? If not: Is this the same case for
> >DiffServ? Does the effect of the IP-packet scheduler get  "disabled" in a
> >similar manner when you run DiffServ over ATM?
> >Yours Sincerely
> >
> >John Tillberg
> >
> >*********************************************************************
> >John Tillberg
> >Telia Research AB      Phone:  +46 8-713 82 07
> >Wireless Solutions     Fax:       +46 8-713 81 49
> >Vitsandsgatan 9                Mobile:  +46 702-43 43 90
> >SE-123 86 Farsta       Email:  John.E.Tillberg@telia.se
> >Sweden

_______________________________________________
Diffserv-interest mailing list
Diffserv-interest@ietf.org
http://www1.ietf.org/mailman/listinfo/diffserv-interest