Long-haul problems - connections stuck in slow start

[Garrett Wollman]

In article
<CAGE5yCpK7N1BwzGCXmx=8M4Q4cu9iUKo4Za-92gRwnkR=HvAiA at mail.gmail.com>,
Peter Wemm writes:

>I'm looking at the duplicate acks in this trace from each end..  any
>clues where they're coming from?

My original suggestion was going to be that they're probably a result
of your packets getting reordered because of that path instability you
pointed out.  10 ms differential delay is much higher than the
interarrival time should be, so you likely have a burst of packets
headed down the slow path, followed by a burst of packets down the
fast path, and the later packets arrive at the destination first,
causing a burst of dupacks until the slow packets catch up.

But that doesn't seem to be born out by your traces.  Do you have TCP
RX offload on this hardware?  Could it be sending the dupacks without
looping them back to BPF?  (I haven't looked at our TCP offload code
so I have no idea if that's even possible.)

-GAWollman