[j-nsp] Redistribute Connected

[Richard A Steenbergen]

On Tue, Nov 25, 2008 at 04:08:45PM +0800, Mark Tinka wrote:
> Setting the NEXT_HOP self attribute, I believe, is good 
> practice, if your intentions aren't whether Arbor or some 
> such software will work correctly or not.

The problem with setting next-hop self on everything is you lose some of
your traffic engineering controls that you would otherwise have if you
knew the next-hop all the way through to the edge. For example, a
classic TE trick is to put a target interface into your IGP as passive
and then set a cost on it, which makes that one interface appear to have
a higher cost than it might otherwise normally have. If you set your
next-hop to self, you lose the ability to do per-interface TE like this. 
Plus if you have a complex network with many potential exits it is darn
handy for troubleshooting if you can see the final destination interface
of a route from anywhere in the network, not just the destination
router.

As with anything, it is a question of scale and what works best for your
network. Personally I find that it makes sense to carry /30s for peering
and transit edges in my IGP, but to do NHS aggregation of customer
aggregation router where such TE is less likely. YMMV, but its always
smart to understand the design considerations and make your own decision
for what is best in your own network, rather than just taking one
person's word as gospel.

Or just redistribute it all into your IGP and let god sort it out. Seems 
to work for some people. :)

http://www.digivill.net/~binary/ccie/levelled3.jpg

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)