First, I have a question about Emmanuel's presentation.
https://datatracker.ietf.org/public/meeting_materials.cgi?meeting_num=68

On the slide that presents simulation results for 60 to 100 nodes,
it says velocity = 3 m/s.  Is this really the maximum node speed?
Or should it be 30 m/s?  I think that 0.934 is a low delivery
ratio for 3 m/s.   Also, what Hello Interval did you use for this
simulation?

I think it is ironic that nobody commented that INRIA's solution is not
robust.  Let me explain.  Chandra's draft uses the pushback timer and
non-active ORs (which I will call backup MPRs) to achieve robustness.
About two years ago, several people from Cisco criticized OSPF-MDR for
not being robustness enough, since its Backup MDRs only guaranteed
biconnected coverage, whereas any router can be a backup MPR.
(I always claimed that one must limit the number of backup relays to
achieve scalability, and later Cisco did this by selecting parameters in
a way that ensured only a small percentage of backup MPRs would fire.)
Simulations consistently showed that OSPF-MDR was as robust as OSPF-OR,
implying that biconnected coverage is sufficient (along with the ability
to respond quickly to topology changes).

So it is ironic that is that nobody commented that INRIA's solution is
not robust, even though it does not use *any* backup relays, nor does
it propose any sort of redundancy in relays or adjacencies.
INRIA's is the ONLY proposal that doesn't try to achieve robustness
through redundancy.  (Maybe they plan to add this later?)
This may partly explain why MPR-OSPF achieved a much lower
delivery ratio than OSPF-MDR in the simulation results I posted
earlier, but I suspect that the choice of LSAs and the fact that MDRs
are self-selected (thus allowing faster response to topology changes)
are also factors.

Now, I want to comment on the similarity between MPR-OSPF and OSPF-OR,
since Acee commented at the meeting that the two proposals are very
different.  My point is that the two proposals are easily mergeable,
and that the only difference in *corresponding* methods are the methods
for adjacency reduction.

The three differences below are methods that are used by one of
the proposals, for which a corresponding method does not exist in
the other proposal, but can be used by the other proposal:

1. For robustness, OSPF-OR uses backup MPRs (non-active ORs),
but the same technique can easily be used by MPR-OSPF.
(MPR-OSPF does not yet have *any* method for achieving
robustness through redundancy.)

2. MPR-OSPF uses multicast LSA retransmissions, and this can
easily be used by OSPF-OR.

3. MPR-OSPF uses path-MPR-based LSAs, and this can easily
be used by OSPF-OR.  In fact, OSPF-OR does not yet propose
any method for LSA reduction, so by default this is the
method that would be used.

Now, I hope you see what I mean.
I guess the question is why the two MPR teams cannot get together
and agree on a single solution based on MPR flooding.
Based on my points above, this should not be very difficult to do.
But of course, if they cannot agree, then the two MDR proposals
cannot be merged. I was just trying to encourage such a merging.

Richard


_______________________________________________
OSPF mailing list
[email protected]
https://www1.ietf.org/mailman/listinfo/ospf

Reply via email to