[Mongrel] Determining ideal number of Mongrels for an app?

Zed A. Shaw zedshaw at zedshaw.com
Thu Nov 30 00:22:03 EST 2006


On Wed, 29 Nov 2006 10:08:38 -0500
"Wayne E. Seguin" <wayneeseguin at gmail.com> wrote:

> 
> On Nov 28, 2006, at 15:49 , Zed A. Shaw wrote:
> > Many folks pointed you at the semi-official tuning documentation.   
> > 20 is totally over kill and will hurt you more than you know.  I  
> > find it's about 2-4/CPU depending on the amount of ram you have and  
> > the speed of your CPU.  BUT YOU SHOULD TUNE EVERY APP EVERY TIME NO  
> > MATTER WHAT YOU CHANGED.  Don't believe anything as gospel, always  
> > test it yourself.
> 
> Zed,
> 
> Do dual-core chips count as two CPU's for this calculation?

Do believe anything I said above.  Find out your own optimal configuration.  The reason is if I say 2-4 people will assume it's gospel and never test their apps.  You might find your app can get better performance with more or less, or you might find it is so slow that really you just need to spread it out across as many as you can.

Test it for yourself, and test it every time a new deployment goes out.
 
> Also what is the recommended amount of ram per mongrel process or per  
> CPU based on your experience?

Again, test it.  You should slam a bunch of your biggest pages with large requests and try to see what it hits as a peak, then set that for you resource calculations.  This is especially true if you use rmagick.

Big thing is to go through your test and development logs and look for huge queries.  Those will kill you every time.  Also, make sure your session clearing stuff is good.  Someone was getting huge memory penalties recently because they had 5.2 million stale sessions in their mysql.

Again, believe nothing, test everything.

-- 
Zed A. Shaw, MUDCRAP-CE Master Black Belt Sifu
http://www.zedshaw.com/
http://www.awprofessional.com/title/0321483502 -- The Mongrel Book
http://mongrel.rubyforge.org/
http://www.lingr.com/room/3yXhqKbfPy8 -- Come get help.


More information about the Mongrel-users mailing list