© 2004
Canonical Ltd.
•
Terms of use
•
Data privacy
•
Contact Launchpad Support
•
Blog
•
Careers
•
System status
•
0e1f616
(Get the code!)
I found this thread helpful, so I thought to add my experience.
In short, I have a dual band wifi router/DSL modem (Arris BGW210-700) which seems to mess with some traffic moving between devices connected at 2.4GHz (my SSH server) and 5GHz (my ssh client). I can avoid this by forcing the use of 2.4GHz by both devices.
The symptom I see is the same random stalled SSH sessions as the reporter. In my case, only about 1 in 20 attempts succeed. Adding the various CLI arguments mentioned seems to change the probability of a stall a little, but none eliminate it.
Running a packet capture on the client machine with wireshark, I see that the stall is followed by a frame labeled "TCP Spurious Retransmission" from server to client, and then some "TCP Dup ACK" from client to server. The frame being resent has a length of only 518 bytes, well below the 1500 byte MTU.
I could successfully 'ping -s 1458 <ip>' in both directions. Wireshark confirms that 1500 byte frames were being sent. Still, I tried changing the MTU on both machines to first 1400 and then 1200. This reduced the chance of a stall to the point where SSH was almost usable.
I was puzzled at this point. I suspected the wifi router as I previously had these machines working through a different (older) router, but wasn't sure how the router could be involved between two local devices. Eventually I realized that the router was bridging traffic since the two machines were connecting to different radios.
I disabled the 5GHz radio on the router to force the client machine to 2.4GHz. At that point 20 of 20 connection attempts succeeded.
Running the client machine (my laptop) at the lower bit rate isn't a permanent solution for me. I doubt I'll make any head way with the router though. sigh... wonderful closed firmware.
I'll also mention a couple of other things I tried which made no difference.
Adding "UseDNS no" to the SSH server config changed nothing.
Disabling the offloading features of the server NIC with ethtool also changed nothing.