Multilink X.25 (was Re: [e2e] TCP design descisions)

[Spencer Dawkins]

--- Jon Crowcroft <Jon.Crowcroft at cl.cam.ac.uk> wrote:
> 
> i'd be more interested in more hard to find design debates
> like 
> 
> iii) wasnt there a multi-link X25? did anyone ever build it
> and if so how well did it work?

I know we built this at Northern Telecom (at the time - now
Nortel Networks) on our DV-1/DNS/{several other names} products
in the late 1980s, and I THINK I'm remembering that we used it
to connect to DMS-250 interexchange carrier switches. The issue
was that we were doing real-time billing transfers that could
overrun 56 Kb links (the fastest we had at the time - how quaint
this seems now), and queuing worked until we started to have
rental cars with rented cell phones, so we HAD to be able to
transfer all the billing records before the customer finished
checking out at the car rental desk (else you never saw them
again!).

I know the implementation "worked", but I remember it at a
fairly low level of sophistication, where we were assuming that
all the links had the same path characteristics and just
round-robined (is there a past tense?) over the link group. Not
sure if we ever got to selecting outgoing links based on queue
depth or not - the environment was so constrained that we could
just tell a telco "you need to upgrade all four links", rather
than trying to solve the general interface selection problem for
mixtures of link speeds and RTTs.

Before too long, we put 10BASE-T cards on the DMS-250 switches,
becauseso many of the billing servers ended up co-located with
switches.

And the DV-1/DVSes actually were fun little boxes - we just
based them on proprietary networking protocols back in the early
1980s and then punted them instead of flipping them to IP-based
protocols in the late 1980s. So close. Oh, well!

Spencer