On 2016-01-21, Stuart Henderson <s...@spacehopper.org> wrote: > On 2016-01-21, Steve Shockley <steve.shock...@shockley.net> wrote: >> A while back [1], I posted a question asking about timeout issues using >> Openup (or any transfers really) to work through a Websense proxy. >> Later, I had problems with Smokeping on OpenBSD showing ~50% packet loss >> going through the proxy. After far too long staring at debug logs and >> packet traces, it turned out that the proxy OS (CentOS) simply wasn't >> passing the traffic through to the proxy. >> >> I found a description of a similar problem on Server Fault [2]. It >> turned out OpenBSD was sending two SYN packets with timestamps (which >> were dropped by CentOS), then sending a SYN without a timestamp (which >> was successful). Setting sysctl net.ipv4.tcp_timestamps=0 on the CentOS >> proxy worked around the problem. >> >> So, what went wrong here? Was the OpenBSD timestamp "too random" for >> CentOS 6.7? Or is there some other issue, and I'm just masking it by >> disabling timestamps? > > Removing timestamps will kill performance unless it's on a slow line. > It gives a good clue though - try this (on the centos box) instead: > > http://stackoverflow.com/questions/8893888/dropping-of-connections-with-tcp-tw-recycle > >
Better reference. http://vincent.bernat.im/en/blog/2014-tcp-time-wait-state-linux.html
