Le 3 oct. 2015 02:50, "Daren Sefcik" <[email protected]> a écrit : > > So after making the changes (somewhat implied by Cyril) I ran apache bench with 2 concurrent instances of "-n 10000 -c 500 -w -k" and the result on haproxy stats page is: > > pid = 18093 (process #1, nbproc = 1) > uptime = 0d 2h55m08s > system limits: memmax = unlimited; ulimit-n = 100043 > maxsock = 100043; maxconn = 50000; maxpipes = 0 > current conns = 2235; current pipes = 0/0; conn rate = 39/sec > Running tasks: 1/2252; idle = 85 % > > and response times from the client are unacceptable, 15-20 seconds or longer. once the apache bench tests finish and concurrent conns go down to a few hundred or less the client response times are normal and quick. Not scientific but during the long wait on the client the browser reports down in the bottom browser bar "waiting for socket..." or "waiting for proxy tunnel..."
Hi, How is the system during stress ? Here are a few things I would have checked in similar situation. Is there an accept filter Is there some kernel messages Load, swap usage, disk space During stress : Is there more sys/interrupt than user cpu usage Link saturation Packet lost Suboptimal firewall rules : replay stress packet filter unloaded. Regards Joris > > TIA for any further help anyone can provide, I really would like to get this figured out.
