Insanely poor IPv6 performance, I'm stumped

[RootWyrm]

Okay, so. I have a 30/10Mbit setup here and I'm using Tunnelbroker because my ISP's official stance on IPv6 is "maybe 2016 but probably 2017." (I'm serious - and so are they, sadly.) Obviously I've got bandwidth to spare and then some, and I know my way around a network quite well.

But I'm getting just absolutely abysmal performance from my tunnel. How abysmal? 60Kbit/sec best case with latency in the 1800ms+ range. I've poked at the MTU repeatedly despite no frag indications, with no luck. I've double checked the gif(4) code to make sure it's not an issue there, nope, it's clean. I've tried two entirely different systems and two entirely different IPv6 tunnels. Both exhibit the exact same symptoms and behavior.

I'm honestly at a loss at this point as to what's going on (and held up by a registrar issue) to cause such horrid performance. I can't find anything to explain it, and IPv4 tests clean. There is no DPI going on and I confirmed the DOCSIS configuration is clean and not interfering either.

MTUs I've tried just in case include basically all of them except 1280. I can force frags, but in normal operation where performance is worst, there are zero frag problems. I thought I was hitting an issue due to VLAN MTU (802.1q in a 1500, silly me for overlooking that initially) but confirmed I don't have frag issues anywhere on the network.

Here's the complete network diagram including provider transit path.

pfSense (gif(4)) -> Juniper EX2200 (no l3, yes vlan) -> DOCSIS -> Arris C4 CMTS -> Cisco Catalyst 3700 w/IP -> Cisco Catalyst 6509 w/IOS -> TEH INTERWEBS!

Any ideas? Seeing it with Ashburn and Chicago both.

[kasperd]

Any ideas? Seeing it with Ashburn and Chicago both.

Take a close look on the roundtrip time of each hop along both the IPv4 paths to each tunnel server as well as an IPv6 path rough each to some well connected IPv6 network. Personally I find the mtr util to be a useful traceroute tool for such purposes.