[j-nsp] Summarize Global Table

[Phil Mayers]

On 10/25/2011 05:21 PM, Brad Fleming wrote:
> Are there any tricks to auto-summarize the contents of a RIB where the tributary routes are not originated locally?
>
> Example:
> Input routes:
> prefix: 1.0.0.0/16, next-hop: 5.5.5.5
> prefix: 1.0.1.0/16, next-hop: 5.5.5.5
> prefix: 1.0.2.0/16, next-hop: 4.4.4.4
> prefix: 1.0.3.0/16, next-hop: 5.5.5.5
>
> Consolidated, Installed routes:
> prefix: 1.0.0.0/14, next-hop: 5.5.5.5
> prefix: 1.0.2.0/16, next-hop: 4.4.4.4
>
> Basically a way to consolidate total number of prefixes entering the FIB.

This (aggregate routes on next-hop) comes up now and again, and the idea 
seems to be controversial for some people. Various criticisms are cited 
- unpredictability (a single route change could result in a large jump 
in FIB use), performance implications, and fairly limited improvement 
(the thread cited below reckons 30% mean, 10% worst case, 45% best case, 
depending on connectivity topology)

Specifically, it's been discussed on the cisco-nsp mailing list, where 
Cisco employees have said "we tried it, it's not that great, we're not 
convinced" (although this was a while back).

Personally I'm not sure I'm convinced; it seems to me that it ought not 
to be hard, and can hardly be worse than what happens when you overflow 
the TCAM. But it's not a new, or uncontroversial, idea AIUI. See e.g.

http://puck.nether.net/pipermail/cisco-nsp/2007-March/039192.html

Cheers,
Phil