Edgecast network issues seen in combination with lowered MTU

[Owen Roth <owen () impulse net>]

Hello,

I had an issue with lowered MTU through a portion of my network, and besides the expected impact, some clients were 
unable to access resources either directly hosted or indirectly served content by EdgeCast Networks (had to look at 
traceroutes and view source to determine). I also found nodes that weren't pingable, and didn't appear to have ICMP 
unreachables available but no directly impacted end-to-end lowered MTU was visible.

It seems to be a PathMTU issue or a weird unidirectional MTU issue (?). I have fixed the lowered MTU, and so restored 
connectivity to affected users -- this is not a current problem but still an incipient one. I consulted with EdgeCast 
NOC in this matter, but it is hard to demonstrate a <1% problem when it isn't apparent from every other network. Or 
they may have contracted clue-immunity, it's hard to tell.

I've used MTUroute/TCProute and "ping a.b.c.d df-bit size 1500". That did NOT show an issue, 1500 bytes were pingable 
from the customer end and edge router. Given that I know it was an MTU issue, as that's how I patched and then fixed 
it, is there another tool that would detect whatever this is?

Has anyone else seen this compound problem with EdgeCast Networks?

Is "ICMP unreachables" ON to NOT break PMTUD per RFC, still best practice or has that drifted?

Regards,
Owen