The MPLS WG Archive

[Eric Gray <eric.gray@sandburst.com>]

Chris,

    Unless the drain rate at each of your egress points exceeds the
capacity of your core, you don't achieve any particularly good
guarantees if all you do is police at the edge.  In fact, what you
might see is worst case behavior where a lot of traffic makes it close
to an egress point and is dropped as a result of local congestion AFTER
having already consumed core resources in getting that far.

--
Eric Gray

You wrote:

> On Mon, Jul 30, 2001 at 08:33:08PM +0300, Juha Heinanen wrote:
> > fred,
> >
> > i have no problems whatsoever with diffserv support in regular 801.1q
> > ethernet switches that allow me to map the user priority field to wfq
> > queues.  rpr doesn't offer this capability.
> >
> > phb stands for "per HOP behavior".  of course it is possible to support
> > diffserv only at the edge, but that doesn't lead to the most efficient
> > use of the bandwidth.  it is the same as using cbr vc to connect routers
> > together over atm backbone.
>
> Err, lets set the perfect world aside for a moment.
>
> If your core has a specific capacity, and you meter, classify, policy at the
> edge, don't you have guarantees which enable differentiated services?  Sure,
> diffserv everywhere would be cool.. but, also $$$ and how real is this real?
>
> You can drive diffserv as far as you want, and theoretically anything's
> possible, but at some point you reach the point of diminishing returns..
>
> Budgetary realities do exist.
>
> Flame away.
>
> Cheers,
> Chris
>
> --
> Christian Kuhtz <ck@arch.bellsouth.net> -wk, <ck@gnu.org> -hm
> Sr. Architect, Engineering & Architecture, BellSouth.net, Atlanta, GA, U.S.
> "I speak for myself only.""