[EMAIL PROTECTED] escribió:
> 
> We have done some performance measurements, running JetSpeed on RS/6000
> machines with one and with four processors. The 4-way machine's processors
> are running about twice as fast as the 1-way's. Surprisingly, performance
> seems to increase proportionally with  the speed of the processors, but
> seems not to improve with the number of processors. Did anybody observe
> similar behavior ?

Yes. I observed similar behaviour under 2-way with Linux Sun jdk 1.3
Server Hotspot.

In my tests, response time grows about 50% between 1 and 2 concurrent
requests. From there on, response time growm linearly with the number of
concurrent requests. With "perfect" concurrency, time should be the same
between 1 and 2 concurrent requests.

> 
> A potential explanation for this behavior might be synchronized statements
> in code that is called directly or indirectly from the service-method of a
> servlet. Are there known problems in this area ?
> 

I have not identified any such execution path. Potential problem areas
are the factories and other singletons, where look up must be performed
for components ( getPortlet(), ...) and also the database (I'm using
Hypersonic for my tests). 

Which DB are you using?

The logging is a big potential bottleneck, since the threads fight to
get a lock on the logger stream. We are logging maybe 20 lines (=20
locks) per request.

Do you have profiling data? The profiler does not work under linux
hotspot :(


--
--------------------------------------------------------------
To subscribe:        [EMAIL PROTECTED]
To unsubscribe:      [EMAIL PROTECTED]
Search: <http://www.mail-archive.com/[email protected]/>
List Help?:          [EMAIL PROTECTED]

Reply via email to