Odd "kernel: TCP: time wait bucket table overflow" vs Bitvise?

[vibroverbus]

Just a bit of an FYI here really but a few times in last week I've had massive time_wait overflow messages overwhelming my logfiles.

Most threads on that error talk about DOS attacks and whatnot, but looking into it, the open time_wait connections all were from my main admin PC and appear to be related to having left Chrome browser SSL GUI windows logged in, and Bitvise open. Seems like something to do with the automatic reconnects after the router went down for nightly reboot. As soon as I restarted Bitvise, it all went away. However I did tweak my TCP Timeouts down a little bit, and that should help as well hopefully if that happens in the future.

If anybody has any other suggesties let me know.

 

[CooCooCaChoo]

Possibly increase time wait buckets as well?

 

[vibroverbus]

Possibly increase time wait buckets as well?

Yeah, my feeling was, better to close the bad sockets a little earlier, and not just make a bigger bucket of bad connections. The only knock-on problem should be a connection with an actual load of latency and/or some weird a slow application behavior expecting the connection to still be open. Reading up on time_wait it seems to me that a full 2xMSL is a pretty huge cushion for modern latencies... Some threads have suggested halving the number (aka 60s) but I went with 90s and have been trying to recreate the problem by leaving a ton of browser and terminal windows open overnight and it hasn't reoccurred. A minute and a half still seems like ridiculously plenty of time to keep a socket in close_wait, but, i'll keep an eye out for weird behaviors...