On 08/06/11 13:19, Sam Wilson wrote:
>> So, I was half worried about there being a raft of complaints from
>> people with broken 6to4 implementations on their desktops today. So far
>> we've had not a single call about it as far as I can tell.
>>
>> Has anybody had any bad fallout at all?
>
> Nothing obvious here at an almost-entirely-IPv4 site. We were expecting
> some calls from VPN users who might have tripped over dodgy IPv6
> connectivity on their remote networks, but nothing has reached us so
> far.
>
> On the other hand<http://hide.dnsalias.net/aaaa/worldipv6day.cgi>
> (referenced on [log in to unmask]) suggests there are PMTUD
> issues between it and www.imperial.ac.uk; the other three[*] JANET sites
> on the list are OK.
Depends on what it means and how they're measuring it;
www.imperial.ac.uk/ redirects to www3.imperial.ac.uk/ and there probably
are PMTUD problems to www3.imperial.ac.uk - we discovered a limitation
of the equipment after 1am unfortunately :o(
However, the TCP MSS should be clamped to 1400 for all SYN/SYN&ACK
packets so it should be working for "actual" traffic.
It's not an entirely satisfactory solution, but analysis of the traffic
hitting the IPv6 webserver indicates very, very few people are getting
through the MSS clamping and still having problems (mainly Sixxs
tunnels, with a default MTU of 1280 - I should probably lower the clamp)
TBH, learning this in itself was almost worth the effort from our PoV;
the existing of significant islands of connectivity with both lower MTUs
(1480, 1280) that are not themselves MSS clamping is interesting, given
the history of IPv4 PMTUD and tunnelling.
|