[nsp] l2tp-cpu load

[Hank Nussbacher]

On Fri, 28 Nov 2003, Andy Furnell wrote:

The 16,000 number probably came from here:
http://www.cisco.com/application/pdf/en/us/guest/netsol/ns242/c666/ccmigration_09186a008015d4d2.pdf
It refers to the NPE-G1 

-Hank

> On Fri, Nov 28, 2003 at 11:05:26AM +0000, H S wrote:
> >
> > Hi !!
> >
> >      I have a Cisco 3640 and have set up an l2tp tunnel. I wonder how 
> > many simultaneous sessions will it support. Currently, I have an average of
> > 16 sessions and 26 % of CPU load. The traffic through the interface is not
> > much (1 Mbps) and (400 pps) I?m planning to change the 3640 by a 7204
> > (NPE-400) but I would like to know it?s limitation on this issue.
> > Could anyone advise me?
> >
> > Thanks in advance!!!!
> >
> > Regards
> >
> > Hugo
> 
> Hi Hugo,
> 
> I had an NPE-400 terminating ~1500 l2tp dial sessions at around 45% CPU...
> This box was also handling a couple of PA-MC-E3 and a PA-MC-STM1 with a total
> traffic load of about 35mbit/sec and ~15kpps. This took some messing around
> to get CPU usage below 60%... PPP compression is right out the window,
> and vpdn ignore udp checksum in your global config will reduce CPU usage by
> another 10-15%. The only problems we ran into were with virtual-access
> interfaces not being re-used - the only solution for this at the time was
> to reload the machine every month, but this bug may well be fixed in a more
> recent IOS release. As far as IOS releases go, we had the best luck with
> 12.2(8)T, although I've heard good things about11T and 15T, as well as 12.3
> mainline.
> 
> Remember the number of tunnels will affect performance much more than the
> number of sessions... if you can aggregate the l2tp tunnels so only one tunnel
> is presented to your LNS you'll get much better performance than if you're
> forming hundreds of l2tp tunnels to hundreds of LACs with a similar number
> of active sessions.
> 
> If the cash is there you'll have get much better performance from an NPE-G1.
> >From speaking to people using them as a broadband LNS they seem to be able
> to take an STM-1's worth of traffic in about 5500 sessions without breaking
> a sweat :)
> 
> A
> 
> --
> Andy Furnell
> andy at furnell.org.uk
> _______________________________________________
> cisco-nsp mailing listcisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
> 

Hank Nussbacher