[cisco-bba] PPPoVPDN scability

[Benoit GRANGE]

Hello all,

> While I haven't tried this specifically, my guess is that you are
> punting packets from the PXF path to the RP by using the 
> tcp-adjust-mss
> command.  The NSE1 is a pxf enabled platform and the 
> subsequent scaling
> claims are based on using the pxf for packet forwarding.  If you are
> forced to use the RP (which is there for punted traffic that 
> is control
> plane oriented) as the forwarding plane, you can see a reduction in
> performance.  

we are running with a similar config on 7401 and we can load the box until it reaches 95% CPU with 2400 L2TP sessions and 125 Mbit/s (30 kpps) outbound to our telco. This is very different from your data (50% cpu at 30 mbit/s).

tcp-adjust-mss does not cost much because the pxf seems to punt only TCP syn segments... not too many of them.

Benoit