On 8 Jan 2004, at 15:59, Sylvain Wallez wrote:
Berin Loritsch wrote:
Just a point of further clarification with Fortress:
The Fortress pool will continue to grow under heavy load, as there is no
true maximum. During periods of light load, the pool is then purged to
more reasonable levels. There is a process that runs periodically to
detect these things. When the pool size is below a certain threshold (i.e.
under such heavy load that there are no more instances available) the
background process will create more instances. The pool will still create
what it needs on demand--but this background process will create more than
one instance at a time. When the pool size grows beyond that threshold
(i.e. the load is down and all the instances are being returned), the
background process will start disposing of the excess.
Cool behaviour. This will certainly make it a strong performance gain, and removes the most difficult part of Cocoon's performance optimization which is sizing pools.
I *hate* sized pools.
Ah, Berin, what strategy algorithm are you guys using for size management?
When writing the thread pool for jserv (god, 6 years ago), I had the feeling that the optimal size of the pool was when the average size equaled its standard deviation but never came to prove it statistically... is the pool strategy pluggable in fortress?
-- Stefano.
