Hi Jens I knew of the 7 thread limit of the DB connections but I wanted to make sure. The tests are nearly finished and I have bad news. As it appears the values for -t and -i are extremely dependant on the type of hardware used and the type of coding (mainly static or manny loops,.-..). E.g. I have noticed, that when using 20 concurrent users the max time for an answer was generally lower when using -t 3 than when using -t 7 (2%-28% faster)! This was a totally unexpected result. As I said before this seems to be a really individual thing. This means that I have to do the same tests on my production system (taking it down for that time), which means LONG nights :-( Now I'm going to test the effect of the Object Cache and -if time allows- the RAM & other caches of Zope. I hope that the performance curve there is a little more predictable.... ;-) regards Oliver Erlewein
On Thursday 27 Jun 2002 12:09 pm, oliver.erlewein@sqs.de wrote:
E.g. I have noticed, that when using 20 concurrent users the max time for an answer was generally lower when using -t 3 than when using -t 7 (2%-28% faster)! This was a totally unexpected result.
Was this keeping a fixed size for the object cache? A fairer test (arguably) would be to increase the object cache size by a factor of 7/3. That should give both tests roughly the same *total* memory usage. (btw, all my production Zopes use -t2)
Now I'm going to test the effect of the Object Cache and -if time allows- the RAM & other caches of Zope. I hope that the performance curve there is a little more predictable.... ;-)
Object Cache performance should be very predictable in Zope 2.6.... the new activity monitor in Control Panel makes it very easy to tune this parameter. In all earlier releases the Object Cache performance was far from predictable.
participants (2)
-
oliver.erlewein＠sqs.de -
Toby Dickenson