[c-nsp] PPP Multilink, inverse of dialer load-threshold for T1 interfaces

[Oliver Boehmer (oboehmer)]

rcheung at rochester.rr.com <> wrote on Friday, March 10, 2006 7:03 PM:

> Hi, I have a bit of an unusual question. We have a PPP Multilink
> bundle consisting of 6 T1s, comprising interface Multilink1. Two of
> them have a second set of PVCs to another
> location, mapped to interface Multilink2. At times, the load from
> these two PVCs can affect traffic on Multilink1.
> 
> Is there a way to configure PPP multilink to switch traffic between
> link members based on the proportion of current traffic existing on
> the links? So if there's high traffic on
> Multilink2, to have more traffic distributed among the 4 other T1s in
> Multilink1 that are not shared?
> 
> PPP BAP appears to be only for dial/isdn interfaces... not sure if
> there's an equivalent of a T1. I guess I'm looking for the inverse of
> the dialer load-threshold for T1 interfaces.

I don't see any easy solution to your problem except to limit the
throughput over the PVCs on the "shared" T1s so they don't affect each
other. 
Multilink can load-share between the bundle members based on the current
throughput and weight of the members, but you want/need load-balancing
on Layer3 level, ie. between the two different bundle interfaces based
on their load. There is currently only EIGRP which can take the load
into account when calculating the metrics, but this is rarely used.

There is no dynamics (like dialer threshold/BAP) for leased-line
multilink.

You could evaluate OER to load-share over both Multilink interfaces as
it can take delay/loss/load/etc. into account and send the traffic over
the "optimal" path. But since traffic over one path can influence the
other (due to the shared T1s), this could also produce interesting
results..

	oli