Hi,
I was running Apache Cocoon with Tomcat 5.0.28 and a Sun 1.4.2 JVM.
Unexpectedly the system was falling over at a certain load and not
gracefully slowing down. At a certain number of concurrent users it
was returning very strange errors including Java Out of Memory
errors, but only after returning some weird ones!
After doing some intensive testing and not getting anywhere I looked
for help from the Cocoon community and was pointed at the JVM because
the stock for garbage collection is no good for web servers and can
be improved with fine tuning via command line parameters. The
following document has some good points regarding it..
http://java.sun.com/docs/hotspot/gc1.4.2/index.html
I actually solved all my problems by upgrading to the 1.5 JVM which
has far better garbage collection handling and I have found under
regression testing the web system gracefully slows down when things
get tough as opposed to falling over, it also performs much better
than it did under the 1.4.2 JVM.
Hope this helps,
David.
On 15 Feb 2006, at 13:29, Leon Rosenberg wrote:
Hi,
has someone ever estimated the max possible load for tomcat? We are
suffering from very odd behaviour, as long as tomcat/machine is under
certain load it works very well, but then it just "breaks a magical
limit" and the performance breaks down.
Tomcat or the JVM are starting to leak memory (we are running with
mx800M, and are normally at 600+ free memory, but then it goes to
zero). Tomcat begins to throw absolutely ununderstandable exceptions,
like beans not present in scope which are normally there, socket/pipe
errors, unable to deliver 5xx or 4xx error pages and so on. Busy
threads are going high, from 10-20 to some hundreds.
We are running with jdk1.4.2_04, tomcat 5.0.25, linux kernel 2.6.8
(debian) 32 bit on different hardware, intel or opteron 1-2
processors, 32 and 64 bit processors (up to 4 virtual processors). No
tomcat clustering, no app. server. Our loadbalancing is based on CPU
Idle time, so the amount of traffic on each machine is different,
faster machines serving 4 times more then slower. The faster machines
are serving 100-150 requests per second in normal, and approx. 300-400
requests per second in peak times. Avg. request duration is <15 ms in
normal times (the servlet execution, not counting the jsp transmission
afterwards) and 100-200 under high load.
We need some serious performance benchmarks, but those are hard to
find (most people talk about absolutely irrelevant things like
database tuning)...
Any good links?
TIA
Leon
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]