[Iccrg] Explicit feedback

[John Leslie]

Doan B. Hoang <dhoang at it.uts.edu.au> wrote:
> John Leslie wrote:
> 
>> Oscillation is a serious problem in routing. In a research environment,
>> it can be measured and appropriate damping can be calcultated. In actual
>> deployment, oscillation must be avoided.
>  
> There are few problems with carrying additional information in routing 
> messages:
> 1) It takes so many years before a routing protocol is adopted (look at 
>    OSPF)

   That's Somebody Else's Problem.

> 2) Many folks have tried to put QoS parameters in the Link State Vector 
>    so that they can find optimal QoS routing. The QoS optimal routing
>    has been proved a NP-complete problem.

   I don't agree that's the problem we need to solve.

>    Even [if] you do not care about optimality or NP-complete, it still
>    does not work in reality (stability reason).

   We really ought to know by now how to control oscillation!

> 3) Oscillation problem was investigated in the early ARPANET, where the 
>    traffic-sensitive link parameters are updated and reported every
>    1/3 of a second (see one of McQuillan's papers), the result was
>    unstable and loops.

   ... which surprised us at the time, but shouldn't surprise us now!

> That is why we talk about updating every 60 seconds or there about

   Oscillation is a problem of phase shifting, not delay time. It's the
damping inherent in averaging which helps here.

   The question should be _how_ to route around congestion. Clearly,
it's not always possible; but when it is, we could be doing a _lot_
better!

--
John Leslie <john at jlc.net>