Since a week or so I am seeing large http up/downloads completely stalling.
The faster the connection, the longer it apears to take to hang.
In testing between our colocation server (with native IPv6) and our tunneled site, it takes some 30GB for a 100GB http download before it hangs.
rsync over ssh hangs almost immediately.
I downgraded the MTU to 1280 at both sides, but it makes no difference.
The problem does not appear to be with HE, because I am seeing the exact same thing in a private 6in4 tunnel between this colocation server's IPv4 address and our IPv4 endpoint. So somehow, our ISP must be doing something wrong.
How can I test what my ISP is doing wrong? I don't see any MTU issues when I run tracepath over the IPv4 route to the HE endpoint:
$ tracepath 188.8.131.52
1?: [LOCALHOST] pmtu 1500
1: wisper-gw.macroscoop.nl (184.108.40.206) 0.775ms
1: wisper-gw.macroscoop.nl (220.127.116.11) 0.707ms
2: static-51-197-117-93.thenetworkfactory.nl (18.104.22.168) 0.325ms
3: 10.99.9.41 (10.99.9.41) 4.244ms
4: 22.214.171.124 (126.96.36.199) 5.437ms
5: 30gigabitethernet1-3.core1.ams1.he.net (188.8.131.52) 5.338ms
6: tserv1.ams1.he.net (184.108.40.206) 5.221ms reached
Resume: pmtu 1500 hops 6 back 6
What does strike odd is the private IPv4 address in the route. Could that be it?