[c-nsp] strange behavior over MPLS network - remote desktopwon't work

[Terry Baranski]

> On Sun, May 31, 2009 at 8:03 PM, Chris Hale wrote:
> 
> I'm using PA-FE-TX as you guessed.  I have enabled the larger 
> MTUs for MPLS
> with mpls mtu 1538 as per others who have successfully used this.
> 
> I'm using one of my J2320's (CE router) and found this:
> 
> ping do-not-fragment detail size 1473 192.168.3.254
> PING 192.168.3.254 (192.168.3.254): 1473 data bytes
> ping: sendto: Message too long
> 
> ping do-not-fragment detail size 1472 192.168.3.254
> PING 192.168.3.254 (192.168.3.254): 1472 data bytes
> 1480 bytes from 192.168.3.254 via ge-0/0/3.600: icmp_seq=0 ttl=62
> time=12.598 ms
> 1480 bytes from 192.168.3.254 via ge-0/0/3.600: icmp_seq=1 ttl=62
> time=14.612 ms
> 
> What does that indicate to you?  1472 + VLAN tag plus MPLS < 1500?

When you ping from a Juniper, the packet size that you specify doesn't
include overhead.  So once you add IP (20 bytes) and ICMP (8 bytes) to the
above, a ping of 1472 is really a ping of 1500.  Therefore 1473 wouldn't be
expected to work if all links on the path have an IP MTU of 1500.

Which CE is the above ping from?  Is it testing the end-to-end path that
isn't working?  If so the MTU seems fine (now).  If RDP still isn't working
maybe you can post the capture.

-Terry