The following reply was made to PR general/843; it has been noted by GNATS.

From: Marc Slemko <[EMAIL PROTECTED]>
To: Nathan J Kurz <[EMAIL PROTECTED]>
Subject: Re: general/843: StartServers directive works but children immediately 
die
Date: Wed, 9 Jul 1997 10:30:07 -0600 (MDT)

 
 On Wed, 9 Jul 1997, Nathan J Kurz wrote:
 
 >  Why are processes replenished at only one per second?  It would seem
 >  that a deficit should be cleared up as soon as possible.
 
 Limiting the rate at which servers are created and killed is a good
 thing(tm) because it helps avoid wild fluctuations due to very short-term
 changes in load.  It is arguable that the method of limitation is bogus,
 but it is partly a result of the current design structure.
 
 >  Should the sleep(1) in wait_or_timeout() be a select() instead?  That
 >  would make it much easier to adjust the rate at which processes are
 >  created.
 
 A select() is used there now, and the interval can be changed.  Check the
 1.3 source tree. 
 
 from httpd.h:
 
 /*
  * (Unix, OS/2 only)
  * Interval, in microseconds, between scoreboard maintenance.  During
  * each scoreboard maintenance cycle the parent decides if it needs to
  * spawn a new child (to meet MinSpareServers requirements), or kill off
  * a child (to meet MaxSpareServers requirements).  It will only spawn or
  * kill one child per cycle.  Setting this too low will chew cpu.  The
  * default is probably sufficient for everyone.  But some people may want
  * to raise this on servers which aren't dedicated to httpd and where they
  * don't like the httpd waking up each second to see what's going on.
  */
 #ifndef SCOREBOARD_MAINTENANCE_INTERVAL
 #define SCOREBOARD_MAINTENANCE_INTERVAL 1000000
 #endif
 
 

Reply via email to