[nsp] OSPF routes

[Tomas Daniska]

> i assume even with a big cpu this number of routes will be a 
> reconvergence
> (major dijska running!) nightmare to any system tho? ..

that's why they introduced incremental spf. only the affected branch is
recomputed so this might save lot of cpu (the farther the event from a
node, the better)
 
> > The number of OSPF routes that can exist in a router is
> > quite large, and it does indeed depend on the capacity of the
> > router (memory, cpu) and also on network design (number of
> > externals, rate of network churn).  While I certainly wouldn't
> > recommend doing this, routers are available that can accept
> > full Internet routes via OSPF.

however i believe peter psenak was talking at networkers about _routers_
(not routes) allowed in ranges of thousands in an ospf domain.

the guys writing the ospf code have repeatedly stated that the
well-known numbers are far from reality now, giving current cpu's and
memory.

iirc, i've seen some measurements showing spf times of <30ms for
1000-node full spf; <5ms for 1000-node incremental spf (transit link
flap). there also were some examples for 2-10k node networks, still in
impressive time ranges

--

deejay