[f-nsp] static vs dynamic cam mode on MLX/XMR (turning foundry direct routing off)

[Gunther Stammwitz]

Sorry.. there was a typo.

I actually wanted to ask if anyone is actually using the DYNAMIC cam mode (when
FDR is turned off) on a peering router with lots of different destinations and
heavy traffic load?

I guess that the dynamic cam mode is no problem as long as most of the prefixes
are being routed to the same next-hop because net-aggregate can aggregate the
cam very well.
The question is what is going to happen on an ixp- or transit-router with lots
of different next-hops??

In dynamic cam mode prefixes are added dynamically to the cam when there's
actual traffic directed to them and they are being removed from them cam after
some time when they expire. This process is utilizing the cpu (I guess: the
linecard's cpu???).

We can see on the ironcore and jetcore modules of the old bigiron that this is
working pretty well as long as there are only a few prefixes per second
addressed or as long as the cam can be aggregated.

As the cam on the mlx is now 262144 entries big compared to 4k or 8k on the old
bigiron most prefixes can be kept in the cam. this should work even when the
routing is very diverse, shouldn't it?

Has anyone actually tested the dynamic cam mode on the MLX?? What are your
experiences: and how good or bad is this working and what's the cpu-load? Can a
traffic-limit under certain conditions be predicted?



Gunther




> Is anyone actually using the static cam mode on a peering 
> router with lots of peers, maybe at an IXP???
>