Hi Jennifer,
Yes, it is true that the worst case is not indicative of the average
or common case. However, note that in our paper we also examined how bad
can things be if we assume a very simple topology of parallel links with different
capacities (you can imagine these to be parallel routes). In this example,
as well, OSPF can perform quite bad, in the worst case, only a log factor
of the flow can be achieved with OSPF.
Even in the very simple case of 2 parallel routes one with double the capacity
of the other, it is easy to show that OSPF cannot achieve more that 2/3 of the
flow TE can.
I agree with you that in practice most likely the penalty will not
grow as a function of the network size. However, I expect to see
a significant reduction of the network capacity when we avoid using MPLS
and relay on OSPF weights instead.
Yuval
>
> I'm not sure I entirely agree with this conclusion. The fact that the
> gap can be very high in the worst case does not imply that the gap is
> high in practice -- for "realistic" topologies (whatever that means)
> and traffic patterns. This is the main theoretical issue that I am
> curious about (but not equipped to study myself) -- what aspects of a
> topology tend to make the gap less significant? High bisection
> bandwidth? Reasonable balance between the topology and the traffic
> demands (i.e., not terribly skewed loads)? Relatively low diameter?
> These things seem true in practice and, intuitively, they seem to hint
> that the worst-case gap probably won't be realized in practice.
>
> -- Jen
>
>
-- Yuval Shavitt Dept. of Electrical Engineering - Systems Tel Aviv University, Tel Aviv 69978, Israel Tel: +972 3 640 8659 Fax: +972 3 640 7095 URL: http://www.eng.tau.ac.il/~shavitt
This archive was generated by hypermail 2b29 : Mon Aug 04 2003 - 04:10:03 EDT