[c-nsp] CEF inconsistency with 12.2(25)S1

[Bernhard Schmidt]

On Mon, Nov 01, 2004 at 04:51:33PM +0200, Volodymyr Yakovenko wrote:

Hi,

>  Are these problems only IPv6 related?

So far yes, I had the IPv6 problem about ten times now, and no IPv4
problems (also Fulltable).

I suspect the problems appear when the prefix changes too often within a
short period of time. I have two backbone routers having the exact same
eBGP peerings (through Ethernet). So if a good path gets lost for one
router, it might still see the good path through the iBGP peering
between the two routers and choose this. The other router might see the
same update some milliseconds later and the routing has to change again.

So far I had this problem about ten times with an internal /128 prefix
used for anycast DNS resolver. Two servers have four eBGP-sessions to
the backbone each (two sessions through two different subnets to each
router), so when one server is shutdown you might have a massive amount
of path changes within a second.

The k.root prefix is the first global one I've noticed having that
problem. There will probably have been more of them, but I don't monitor
each prefix.

Bernhard