Does anybody have a working projection, or crystal ball, that can provide a recommendation on FIB size requirements for the next 24 months? Are we expecting the IPv4 table to continue to grow at somewhere around 50k routes a year? I came up with this from eyeballing the graph at http://www.cidr-report.org/cgi-bin/plota?file=%2Fvar%2Fdata%2Fbgp%2Fas2.0%2Fbgp-active.txt&descr=Active%20BGP%20entries%20(FIB)&ylabel=Active%20BGP%20entries%20(FIB)&with=step.
Thanks,
Graham Johnston
Network Planner
Westman Communications Group
204.717.2829
johnstong@westmancom.com<mailto:johnstong@westmancom.com>
P think green; don't print this email.
Hi Graham,
The IPv4 BGP table has been growing by 10% to 15% per year since CIDR.
It appears to be a compounding curve, not linear.
IPv4 exhaustion is a new factor which may or may not impact the next
24 months' projection. There are arguments favoring a slower rate (no
more free pool). There are arguments favoring a faster rate
(fragmentation from address sales). No one has a crystal ball good
enough to know for sure -- the situation is literally unprecedented.
Regards,
Bill Herrin
When will router vendors learn to do even simple aggregation before loading
routes into FIB? It appears that most hardware will have plenty of FIB
space if this was done. Also that aggregated routes are increasing at a
slower pace.
Regards,
Baldur
When de aggregation hit IPv6, with lot of /48
Howdy,
You can get a good reduction with FIB aggregation, but only upwards of
50% or so, and that only with the some pretty costly algorithms. Also,
you tend to get better gains at the cheaper edge nodes rather than the
more expensive core nodes. For now it's more cost effective to leave
the code alone and just double the size of the TCAM.
Regards,
Bill Herrin
> When will router vendors learn to do even simple aggregation before
loading
> routes into FIB? It appears that most hardware will have plenty of FIB
> space if this was done. Also that aggregated routes are increasing at a
> slower pace.
Howdy,
You can get a good reduction with FIB aggregation, but only upwards of
50% or so, and that only with the some pretty costly algorithms. Also,
you tend to get better gains at the cheaper edge nodes rather than the
more expensive core nodes. For now it's more cost effective to leave
the code alone and just double the size of the TCAM.
Regards,
Bill Herrin
There are also features such as Selective RIB Download (or its equivalent
in other vendors) which help out in different portions of the network.
Definitely not applicable to all router types though.