[altq 349] Re: DiffServe setup

[Kenjiro Cho <kjc@csl.sony.co.jp>]

Vangelis Pappas-Katsiafas wrote:
> I'm trying to set up Diffserve in a small test network. I received some
> very useful comments a few days ago but still I have some points that
> are not very clear to me. Basically it has to do with wich interface
> does what. I guess any configuration will work - since altq accepts the
> altq.conf file - but the point is which is going
> to behave best.

It depends on your assumption about how traffic are merged.

> My network is quite simple: one ingress, one core, one egress.
> 
> Ingress:
> It's almost clear from the example in TIPS.txt that the input IF acts
> like the TC (MF classification, shaping, marking) and the output forms
> the queues for the EF and AF classes. (right?)

Yes.

> Core: (in my case simply 1 input and 1 output IF)
> Do we need both IFs to form AF/EF queues? I assume all IFs in the core
> have to have the AF/EF queues. This becomes more apparent when I think
> of a more complex DiffServe network. Then we might need TC on the core
> too.(?)

If you mean that "input IF" has only incoming traffic and "output IF"
has only outgoing traffic, a queueing discipline is needed only on
"output IF".
If traffic is bi-directional, "input IF" is an output interface for
packets in the opposite direction and it needs a queueing discipline.

Whether you need traffic conditioning or not depends on your
assumption about how traffic is merged at this point.

> Egress:
> Again I assume only AF/EF queues in both input and output IFs. Output
> can optionaly clear DSCPs. No TCs are required. (?)

The same rule can be applied to egress routers but you probably have
different assumptions on egress points.

-Kenjiro