On Thu, May 8, 2008 at 12:04 PM, Tom Storey <tom@snnap.net> wrote:
>
> As for the NPE-G1, typically Ive seen around 4000 sessions terminated
> before the CPU really starts getting bogged down. Thats with an ACL
> and rate-limit per user.
>
I can confirm having seen this kind of figure on the G1. The limit really
appears to be CPU and is effected not only by the number of tunnels but by
the bandwidth and packet rate per tunnel. This means that the type of user
and application can have a big impact. Also certain features, like
cache-flow use significant extra CPU and thus reduce the max tunnels.
Due to the fact that in this application the box appears to be primarily CPU
bound we have found the amount of tunnels / traffic scales with the CPU
speed, i.e as you go from NPE400 to G1 and then G2 the number of tunnels (
assuming the same bandwidth and packet rate per client) scales
proportionally to the increase in CPU speed.
We had little success with the MPF features available on the G1.
(
http://www.cisco.com/en/US/docs/ios/12_3t/12_3t7/feature/guide/MPF123T7.html )
Although it seemed to address CPU load to some extent it introduced various
other forwarding issues.
Due to the variables involved there seems to be very little firm
documentation on the subject of actual number of tunnels.
Regards
Neil