[Olsr-users] Sticky gateway

[Juliusz Chroboczek]

Phew, a lot of reactions here.

Henning:

> Yes, theoretically this can happen if the "sticky" threshhold is too large. 
> That's the reason why we suggest a default of 0.3, which is much lower than 
> the cost of a single hop.

By reducing the threshold, you reduce the likelihood of the loop happening.
You don't reduce the time for which it persists once established, which
remains unbounded.

Sven-Ola:

> b) the feature minimizes _gateway switching_(*)

I'm not arguing with the advantages of the feature, which I fully
understand (as a matter of fact, I do have hysteresis in Babel).  I'm
saying that *in a link-state framework*, it has some very significant
issues, and hence should not be recommended unless people fully understand
its flaws.

There's no reason, by the way, why OLSR shouldn't be extended to become
a hybrid protocol, which uses link state for internal routes and
distance-vector for external ones.  See OSPF.

Aaron:

> First of all, I want to ask you to elaborate a bit on your claim.
> Can you give an example?

Yes, I can.  But do I really need to construct an example whenever
I express a doubt about any of the features you're introducing?  I'd expect
the duty of proof to be on the proponent's side.

Consider the following topology:

              3     3     1     2     2
   0.0.0.0/0 --- A --- B --- C --- D --- 0.0.0.0/0
             
Both B and C go through D for reaching the Internet.  Suppose now that it
stops raining around A, and the link costs change:

              1     1     1     2     2
   0.0.0.0/0 --- A --- B --- C --- D --- 0.0.0.0/0

C, which is configured without a sticky gateway, switches to going through
A.  However B, which is using the sticky gateway feature, sticks to going
through D.  You have a routing loop between B and C.

A few points that I make without proof:

  - the routing loop will never be cleared -- it will remain until the link
    costs change again;
  - in the example above, I assumed that B had a sticky gateway factor of
    3; however, by tweaking the values above, you can build a routing loop
    whatever B's stickyness factor as long as it's > 0;
  - in the example above, I assumed that the configuration of your nodes
    was not uniform (B is sticky, C is not); it is possible to construct an
    example where all nodes are configured the same.

Sorry for the lack of proof for the points above, but it's been a hard day.

                                        Juliusz