[c-nsp] MTU over L2TPv3 tunnel

[Oliver Boehmer (oboehmer)]

alex at alsn.be <> wrote on Thursday, February 28, 2008 10:44 AM:

> Hey,
> 
> I have a problem that i just can't resolve. I hope someone will be
> able to help.
> 
> Here's the situation. I have 2 client switches (Catalyst 2400) each
> connected on a separate border router (7200). Between the router is a
> L2TPv3 tunnel.
> 
> When i test the connection (RFC 2544) i get the result i expect
> +/-90Mb/s as long as the packet size is smaller then 1518. Once the
> packet size is 1518 the cpu of one of the two routers gets to 100%
> and the throughput lowers to +/-30Mb/s after about 30 seconds.
> 
> I tested the maximum MTU with a ping and found out it's at 1472.
> 
> I tried to change the ip mtu and mss to correct values (even tried
> lower values) but it didn't help. When i sniff the connection i
> noticed the mss value isn't the value i put one the interface. The
> lowest value i have seen is 1460 (wich should be allright).

Well, you need to increase the IP MTU on the core/uplink interface to
avoid fragmentation (as you've already assumed). You might fail doing so
as you seem to be using an I/O FE port which doesn't support a higher
MTU, but on the NPE-G1 side, you should be able to my increasing "mtu"
on the main interface and setting "ip mtu" on the subinterfaces (if you
want them lower than the main MTU).
enabling PMTU within the PW-class sets the DF-bit, so packets are
dropped if they need to be fragmented.

	oli