can anyone give pointer to documentation that would give good ideas for starting numbers to set for min and max procs for fcgi for different levels of traffic... doesnt have to be entirely precise just looking for some numbers to start from... _______________________________________________ Rails mailing list Rails-1W37MKcQCpIf0INCOvqR/iCwEArCW2h5@public.gmane.org http://lists.rubyonrails.org/mailman/listinfo/rails
> can anyone give pointer to documentation that would give good ideas for > starting numbers to set for min and max procs for fcgi > for different levels of traffic... doesnt have to be entirely precise > just looking for some numbers to start from...The Rails book discusses this stuff in more detail, but I seem to think 10-15 is a good fit if you have a dedicated machine to serve it. The easy way of testing is just by playing with ab/siege, different levels of concurrency, and different numbers of FCGIs started. -- David Heinemeier Hansson http://www.loudthinking.com -- Broadcasting Brain http://www.basecamphq.com -- Online project management http://www.backpackit.com -- Personal information manager http://www.rubyonrails.com -- Web-application framework
* David Heinemeier Hansson <david.heinemeier-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> [0625 19:25]:> > can anyone give pointer to documentation that would give good ideas for > > starting numbers to set for min and max procs for fcgi > > for different levels of traffic... doesnt have to be entirely precise > > just looking for some numbers to start from... > > The Rails book discusses this stuff in more detail, but I seem to > think 10-15 is a good fit if you have a dedicated machine to serve it. > The easy way of testing is just by playing with ab/siege, different > levels of concurrency, and different numbers of FCGIs started.While we''re on the subject, lighttpd seems to have turned off adaptive spawning in the last release. Does that mean that lighty starts at min-procs, ups to max-procs, and sits there? -- ''Everyone''s always in favour of saving Hitler''s brain, but when you put it in the body of a Great White shark suddenly you''ve gone too far..'' -- Prof. Farnsworth Rasputin :: Jack of All Trades - Master of Nuns
On Jun 21, 2005, at 11:45 AM, Dick Davies wrote:> While we''re on the subject, lighttpd seems to have turned off > adaptive spawning in the last release. Does that mean that lighty > starts at > min-procs, ups to max-procs, and sits there?It means set a sufficient min, and if it''s not enough (the clue being a lag or unresponsive hits), then up it and restart. Considering how 1.3.14 fixed so many of these cycling issues were fixed, it''s a great release (the ones where if backend fcgi processes or proxied servers goes down, then lighttpd would cycle, generate a log file at 20GB/min and spiral up to the max memory and then sit there or die; those are ALL gone.) - Jason
On Jun 21, 2005, at 7:10 AM, Sean T Allen wrote:> can anyone give pointer to documentation that would give good ideas > for starting numbers to set for min and max procs for fcgi > for different levels of traffic... doesnt have to be entirely > precise just looking for some numbers to start from...And to go back to this one, on 4 processes (and this is with wordpress.org, our site in textpattern, and we tested our hieraki install), we can comfortably do 30-40 req/sec (and about 500 req/sec with various cachings) at concurrencies up to ~600. Taking it up the number of processes up primarily affected the concurrency. So as an example, wordpress.org on php-fcgi (it''s about same as ruby- fcgi) does 350,000 hits a day on 4 fcgi processes. The yubnub.org site the other day was slashdotted without issue and it was running 4 ruby-fcgi processes ... - Jason _______________________________________________ Rails mailing list Rails-1W37MKcQCpIf0INCOvqR/iCwEArCW2h5@public.gmane.org http://lists.rubyonrails.org/mailman/listinfo/rails
> The yubnub.org site the other day was slashdotted without issue and it was > running 4 ruby-fcgi processes ...Yeah, my experience may be tilted by the fact that we have long-running requests on Basecamp like uploading to a remote FTP server. In any case, I don''t think it''s a good idea to go dynamic on spawning. The only thing you save is a little bit of RAM and you gain a world of issues (like the server being too slow to spawn enough processes and you loosing the cache when they''re killed and continous incurring startup costs etc). -- David Heinemeier Hansson http://www.loudthinking.com -- Broadcasting Brain http://www.basecamphq.com -- Online project management http://www.backpackit.com -- Personal information manager http://www.rubyonrails.com -- Web-application framework
* Jason A. Hoffman <jason-xKtDo/uLHBtl57MIdRCFDg@public.gmane.org> [0655 22:55]:> > On Jun 21, 2005, at 11:45 AM, Dick Davies wrote: > > >While we''re on the subject, lighttpd seems to have turned off > >adaptive spawning in the last release. Does that mean that lighty > >starts at > >min-procs, ups to max-procs, and sits there? > > It means set a sufficient min, and if it''s not enough (the clue being > a lag or unresponsive hits), then up it and restart.do you mean max? That''s what i was seeing, and it was my own fault for being too stingy in the number of procs.> Considering how 1.3.14 fixed so many of these cycling issues were > fixed, it''s a great release (the ones where if backend fcgi processes > or proxied servers goes down, then lighttpd would cycle, generate a > log file at 20GB/min and spiral up to the max memory and then sit > there or die; those are ALL gone.)Don''t get me wrong, I''m loving lighty - I wasted an afternoon trying to get apache doing php-fcgi then cheated and mod_proxied to lighttpd (I have the one IP and a few things that have to be on apache (svn repos etc) ). The reason I''m flapping about the number fcgi children is that I''ve decided to avoid the ulcers that suexec would cause me and just give each user their own lighty instance and mod_proxy to each of them, which nicely isolates their blogs. I just get niggled by the number of idle procs sitting in the process listing. Being pragmatic I suppose the idle ones get swapped out anyhoo, so life is probably too short to worry about dynamic scaling of child pools..... -- ''Domestos is an ideal substitute for Blue Curaco, and far less pricey.'' -- James Francis, East Glamorgan Hospital. Rasputin :: Jack of All Trades - Master of Nuns''Sweet Zombie Jesus!''