[c-nsp] TCP Throughput / MTU problem ? with Cisco 7304 andMPLS VPN's

[Peter Rathlev]

On Fri, 2008-01-11 at 10:48 -0500, Eric Kagan wrote:
> I have never changed a router interface MTU on the 7206's, only the
> switches.  Routers always MPLS MTU xxxx.  That's where I wonder why now ?
> 
> I replied with some of this in the previous response, but the 7206 has
> physical interface MTU = 1500, MPLS MTU 1546 and works fine on over 20
> routers.  Its just the one 7304 that is having an issue.  Currently, I have
> not changed physical interface MTU on 7304, its 1500.  I have tried MPLS MTU
> 1546 (and yes I got the error msg setting MTU above physical can cause
> problems....)
> 
> Jan  8 22:04:47 est: %MFI-3-MPLS_MTU_SET: Setting mpls mtu to 1520 on
> GigabitEthernet1 which is higher than the interface mtu 1500.  This could
> lead to packet forwarding problems including packet drops.

Sometimes the controllers allow larger packets to be sent, but there's
no guarantee. Would there be any problems associated with adjusting
interface MTU up everywhere? Any limits on the data link layer?

> I have not seen any problems with ICMP.  I can ping between the 2 P routers
> connected on the fiber up to 2000 bytes without a problem:
> 
> #ping
> Sending 5, 1480-byte ICMP Echos to 66.203.65.25, timeout is 2 seconds:
> !!!!!
> Success rate is 100 percent (5/5), round-trip min/avg/max = 1/1/4 ms
> 
> Sending 5, 1500-byte ICMP Echos to 66.203.65.25, timeout is 2 seconds:
> !!!!!
> Success rate is 100 percent (5/5), round-trip min/avg/max = 1/1/4 ms
> 
> Sending 5, 1520-byte ICMP Echos to 66.203.65.25, timeout is 2 seconds:
> !!!!!
> Success rate is 100 percent (5/5), round-trip min/avg/max = 1/1/1 ms
> 
> Sending 5, 1546-byte ICMP Echos to 66.203.65.25, timeout is 2 seconds:
> !!!!!
> Success rate is 100 percent (5/5), round-trip min/avg/max = 1/1/4 ms
> 
> Sending 5, 2000-byte ICMP Echos to 66.203.65.25, timeout is 2 seconds:
> !!!!!
> Success rate is 100 percent (5/5), round-trip min/avg/max = 1/1/4 ms

You have to set the DF bit, or else the routers will just fragment and
reassemble. It's an extended ping option.

Regards,
Peter