Hi

Ben & I have made some estimates, based on a future busy = hour for a national network with about 100 PCN-boundary-nodes. We assume that the = case of interest is where the traffic follows this normal pattern and that the = network is (becoming) pre-congested due to a massive surge of traffic outside = this traffic pattern.

Our estimate is that 3/4 of ingress-egress-aggregates will have = traffic of less than 1.5 Erlang. The distribution is long-tailed, for example = there are several ingress-egress-aggregates with > 4000 = Erlangs.

We have made some speculations about how this translates into = traffic on particular interfaces. The topology of the network has each PCN-boundary-node attached to 2 different PCN-interior-nodes; PCN-interior-nodes are more interconnected (> dual-homed). =

So for a particular PCN-boundary-node to PCN-interior-node = interface, it potentially has 50 ingress-egress-aggregates on it; we estimate that actually 25 of these will not have a flow, 10 will have 1 and 15 will = have >1.

For a PCN-interior-node to PCN-interior-node interface, the = topology suggests potentially 300 ingress-egress-aggregates on it; the fractions = are the same as before, ie we estimate that actually 150 of these will not have = a flow, 60 will have 1 and 90 will have >1.

We = re-iterate that these are very rough estimates. But they strongly suggest that = there are likely to be significant numbers of aggregates with very few flows under = nearly all circumstances.

Best wishes,

Phil & Ben

> -----Original Message-----

> From: Lars Eggert = [mailto:lars.eggert@nokia.com]

> Sent: 25 October 2007 13:25

> To: Eardley,PL,Philip,CXR9 R

> Cc: babiarz@nortel.com; pcn@ietf.org; Ruediger.Geib@t-systems.com

> Subject: Re: [PCN] Architecture draft - probing section = & general updates.

>

> On 2007-10-25, at 13:56, ext = philip.eardley@bt.com wrote:

> > Lars, I believe the guess is that there'll be many = ingress-egress

> > pairs with none or very few flows at a particular = moment. Not a

> > random distribution of calls onto all possible pairs. Basically

> > there are many calls between London = & Manchester and very few

> > between truro and Shetland.

>

> I can agree with that. But note that that alone is not a = problem yet.

>

> The problem occurs only when several of these = "empty" ingress pairs

> share a bottleneck *and* when single flows start = simultaneously

> sending across them at a combined rate that will push the bottleneck

> directly into overload.

>

> I don't believe this can be very common. Wouldn't you = provision your

> network such that interior routers could handle at least = one flow per

> ingress/egress pair? In which case there is no issue unless there's a

> failure. And why not let flow termination handle these rare = cases?

>

> > I & Ben will try and extract some numbers from the = [bt] predicted

> > future traffic matrix.

>

> That'd be *very* useful! We're all guessing probabilities = here, and

> data would help.

>

> Lars