Hi Emilio, I tried your parameters, but there was no change in the behavior of the pound-system, still 40%-50% cpu and not even a little more. I even tried a reboot with your parameters in place, but (as it should be) it was the same.
Thank you for your help. kind regards fatcharly -------- Original-Nachricht -------- > Datum: Thu, 9 Feb 2012 08:22:37 +0100 > Von: Emilio Campos <[email protected]> > An: [email protected] > Betreff: Re: [Pound Mailing List] strange performance > Hi fatcharly, Did you do some tunning performance in your operative > system? > > Maybe this parameters could help you: > > Write this parameters in your /etc/sysctl.conf > > fs.file-max = 100000 > net.ipv4.tcp_max_tw_buckets = 100000 > net.ipv4.tcp_low_latency = 1 > net.ipv4.tcp_tw_reuse = 1 > net.ipv4.tcp_tw_recycle = 1 > net.ipv4.tcp_keepalive_time = 512 > net.ipv4.tcp_fin_timeout = 5 > net.ipv4.inet_peer_maxttl = 5 > net.core.rmem_max = 262143 > net.core.rmem_default = 262143 > net.ipv4.tcp_keepalive_intvl = 15 > net.core.netdev_max_backlog = 3000 > net.core.somaxconn = 3000 > net.ipv4.tcp_keepalive_probes = 5 > > After, execute sysctl -p and restart your pound service. Repeat your test > and tell us. > > This TCP/IP tunning help to use less operative system resources and work > like a charm for us! > > I'm very interested on read your feedback > > Regards! > > > 2012/2/8 Joe Gooch <[email protected]> > > > Kemp is purpose built, so I'd expect they've spent a lot of time with > > custom performance tuning. Plus micro-benchmarks never really tell the > > full story. > > > > For the numbers to mean anything you'd need to provide > > 1) Is this SSL or non-SSL traffic? > > 2) How many of these clicks are SSL handshakes - versus sessions that > are > > already established? SSL Handshakes are where the bulk of the CPU load > will > > come from. > > 3) Are these clicks new TCP sessions or are they on a shared > > connection(i.e. HTTP Keepalive) > > > > Recommendations for increasing your number: > > 1) Apply this patch: > > http://goochfriend.org/pound/pound_2.6_SNI_Optimization.patch to prevent > > some possibly unneeded SNI checks in SSL handshakes > > 2) Investigate and change your Threads value. The default of 128 may be > > your limiting factor. Experiment with more or fewer threads to > determine > > the best trade-off of performance and memory. > > > > > > I would bet increasing the Threads value will help in some regard... Let > > me know how it goes please! > > > > Joe > > > > > -----Original Message----- > > > From: [email protected] [mailto:[email protected]] > > > Sent: Wednesday, February 08, 2012 12:02 PM > > > To: [email protected] > > > Subject: [Pound Mailing List] strange performance > > > > > > Hi, > > > > > > we are running a pound 2.6f with a patch v2 by Joe Gooch. > > > When we start to stress our webserver via pound we get ~170000 clicks, > > > while the cpu of our pound-system /Xeon 3,0 GHz 1Gb RAM stays at about > > > 40-50% (top). With the same test-setup but with a different SSL- > > > Loadbalancer (KEMP), we get ~250000 clicks. It seems as if there is a > > > barrier we can´t brake to get more clicks via the pound. Not the > pound > > > or the webservers seems to operate with a high load. > > > > > > Any suggestions are welcome > > > > > > Kind Regards > > > > > > fatcharly > > > > > > -- > > > Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen > Sie > > > mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de > > > > > > -- > > > To unsubscribe send an email with subject unsubscribe to > > > [email protected]. > > > Please contact [email protected] for questions. > > > > > > -- > Load balancer distribution - Open Source Project > http://www.zenloadbalancer.com > Distribution list (subscribe): > [email protected] -- Empfehlen Sie GMX DSL Ihren Freunden und Bekannten und wir belohnen Sie mit bis zu 50,- Euro! https://freundschaftswerbung.gmx.de -- To unsubscribe send an email with subject unsubscribe to [email protected]. Please contact [email protected] for questions.
