Global Lambda Integrated Facility

Subject Why do we need topology exchange?
From Freek Dijkstra <Freek.Dijkstra@xxxxxxx>
Date Thu, 14 Oct 2010 11:35:29 +0200

In his presentation today, Jeroen expressed that while there has been
apparent need for topology descriptions (e.g. for path finding,
monitoring and provisioning), no real adaption has been done by GOLEs
and NRENs.

So let me ask all of you, why do we need topology exchange?

Currently, no GOLEs really distribute topology information (even though
the usual suspects -ESnet, Nordunet, and SURFnet- have expressed their

Is there no need for topology exchange after all?

Right now, all GOLEs use email, pictures, wikis and a lot of smart
people for the operation of their production network. In the last five
years, we as GLIF comminutiy have show that we are successful in
providing network connection accross te globe.

We even show very fancy proof of concepts like yesterday's Dynamic GOLE
demo, but also projects like G-Lambda, and Phosporous in earlier years,
and production application like the e-VLBI network.

However, we may blind ourselves by telling that we now have dynamic
GOLEs. We have not. I don't want to discredit the applications, but the
GOLEs are still manually configured. What is done is that we create an
overlay network for each group (either a project, demo or user
application) on top of the GLIF network. Such overlay network contains
and links and sometimes also dedicated switches or interfaces, which may
be dynamic. However, that does not make the underlying GLIF network dynamic.

So who would need automated topology exchange? The GOLEs? No, since that
is still provisioned by hand (and given that the biggest issue is
dealing with policies and politics that come with scarce resources
typically during demos, this may not change as long our resource remain
scarce). So do the overlay networks need automated topology exchange? We
surely have seen so in demos! Well, arguably they don't need it either.
In most cases, the projects are relatively small, and so is the network
-- typically 5 to 10 nodes only. You can easily just make a nice picture
or hard-code it in software, no need for more dynamics.

Of course, our NREN networks are larger, and do need automated topology
descriptions, but as Jerry pointed out, the issue there is about
automated topology discovery, not topology description or topology exchange.

So, last option -- do we need topology exchange between the overlay
network and the underlying GLIF network? Well, if the mapping is static,
perhaps not. Perhaps it would be useful for monitoring to have a shared
resource description, but that is distinct from topology exchange. (This
has more to do with identifiers, than with topology descriptions)

I think that if we want to move this work forward, we first need to make
clear what the problem is, before we find a solution.

So I'm again asking you: Why do we need topology exchange?

Awaiting your answers,