[discuss] Off topic: BGP-4 growth [was: we need to fix what may be broken]

Sat Apr 19 22:44:25 UTC 2014

Barry,

My last post on this since it is way off topic.

On 19/04/2014 18:25, Barry Shein wrote:
...
> Someone asked what could be done if the routing table were collapsing.
> 
> I suggested one easy thing that could be done: Reduce the number of
> routing table entries by fiat.
> 
> We do that in the core already by limiting the longest prefix so it's
> not completely far-fetched, just extrapolative.

Exactly. When an operator feels that their own BGP-4 table is getting
too close to the current performance limit of their installed BGP-4
routers, they do what they've been doing since 1993: filter long prefixes
more aggressively. The factual evidence is that this has consistently
kept the growth in BGP-4 table size very much sub-linear as a function
of growth in the Internet for the last twenty years. There have been
various approaches to studying this; my own approach is documented at
https://www.cs.auckland.ac.nz/~brian/sqlaw.html .
You can also look for recent talks by Geoff Huston for a different
approach.

The bottom line is that BGP-4 growth appears to be self-regulating,
and the crisis feared in 1992 (RFC1380) and 2006 (RFC4984) has not
yet arisen.

It's a bit soon in the growth of IPv6 to start extrapolating the
graphs, but it seems very likely that the same will apply, perhaps
even more so since IPv6 prefix allocation has been approached
prudently by the RIRs from the start.

    Brian