[c-nsp] Dampening

[David Barak]

--- Brian Feeny <signal at shreve.net> wrote:
<snip>

> One way that seems to have merit is using
> progressive dampening, where
> you penalize a /24 more than say a /8 or /16.  The
> idea being that alot 
> more people
> are going to be peeved if a /8 is not reachable.

I've seen that argument made before, and it seems to
make sense, but it would tend to punish those people
who obtained PI space from ARIN, etc.  

> 
> Also, I believe that no matter what, a route should
> not be dampened 
> unless it has flapped
> at least 3-4 times.  Alot of things go on in the
> operational day to 
> day, where a prefix may flap
> 1, 2 or even 3 times, and that should not be seen as
> a problem.

Yes, although here's the kicker - from Randy's
presentation, if I'm multihomed and shut down one
router, my upstream receives a withdrawal, right? 
After a few minutes, they'll get another add (as they
eventually learn about my other route.  When I bring
the router back up, that's another add right there,
and you're above the 3-flap limit of an aggressive
dampener.

In any case, the original poster was asking about how
to deal with the situation where one of their
providers dampens and the other doesn't (my
suggestion?  get another provider...), and whether
"fast external fallover" is a good idea in that
situation: 

the answer is no, because fast-e-f will tend to
minimize waiting for session timeouts, but will
maximize flapping in the process.  That's not
something which will work well with that provider.

-David Barak


		
__________________________________ 
Do you Yahoo!? 
Yahoo! Small Business - Try our new resources site!
http://smallbusiness.yahoo.com/resources/