[arin-ppml] Does Moore's law help with routing table growth?
(from the wayback machine)
On Sat, Dec 19, 2009 at 12:00 AM, Scott Leibrand
<scottleibrand at gmail.com> wrote:
> On 12/17/2009 7:45 PM, Michel Py wrote:
> Well, back to the original topic, what we need to be concerned with for
> policy is not the relationship between bandwidth supply and demand, but
> the relationship between the growth of the BGP routing table and the
> growth of FIBs to accommodate it. It appears to me that we're doing OK
I think, to be really clear there are two problems, one tends (today)
to track with the other though:
1) size of FIB (in general you can find a way to scale the RIB since
that happens on the much cheaper part of almost all core routers)
2) speed of change of the FIB (RIB -> FIB convergence rate)
FIB size means TCAM or SRAM size
Speed of change is how quickly you can update, and often across a
severely limited pathway, the FIB when a RIB change happens. Today you
stay ahead of that change rate (mostly) and your device's view of the
world is 'converged'. Tomorrow if that change rate is higher than you
can keep up with you will never converge.
> there: router growth seems to be mostly keeping ahead of table growth.
> As long as we don't dramatically increase the rate of growth of the
> routing table, I think we're ok. And AFAICT none of the policy
See the note about change rate, if the current change rate (for
example) 10%/second == 35k changes/second and the pathway you can make
these across can only support 50k/second. You need to be sure that
growth rates won't get you to 500k prefixes before your next upgrade
cycle. Also, it'd be nice if the vendor you choose for that upgrade
didn't keep their current 50k rate-limited pathway intact.