tomcat-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Steve Ruby <st...@rubysolutions.com>
Subject Re: performance
Date Fri, 02 Feb 2001 21:57:10 GMT
Steve Ruby wrote:
> 
> Stefan Langer wrote:
> >
> > I'm following this performance issue with great interest and I was wondering:
> >
> > was the testing for the performance of tomcat based on a single instance of tomcat
running?
> > Did you use loadbalancing?
> >
> > In short I would greatly appreciate more information of the overall enviroment you
tested the load on.
> >
> > Btw are there any sites out there that have benchmarks on them? I know this question
was asked before but I didn't find a satisfying answer yet.
> >
> > Stefan
> >
> > P.S: In a world without fences who needs gates?
> 
> Here's a little curious benchmark I just ran.. maybe meaningless in
> the "real world" but it pertains to my app so it is meaningful to me.
> 
> Hitting a single servlet (turbine based) with maybe 4 queries to
> MySQL generated by hitting the page I get these results
> on a PII400, running SUN JRE 1.2.2 with JIT disabled. I was hitting
> the servlet with 10 threads from JMeter on 300MS delay. The time
> is the average response time of the servlet when then it reaches
> a steady state.
> 
> Tomcat-3.2.1 - green-threads      5.7sec avg (ouch)
> Tomcat-3.2.1 - native-threads     3.4sec avg
> Tomcat-4 -     green-threads      3.8sec avg
> Tomcat-4 -     native-threads     2.3sec avg
> 
> Note it appears that tomcat 4 is a bit faster in this test, and
> definately if I move the Jmeter up to 20 threads hitting a page
> under that tomcat instance with a brower is very bad in 3.2.1 but not
> so bad under tomcat 4. Even at 20 threads hitting it pretty fast
> I can still "use" the site when running under Tomcat4 under 3.2.1
> it seems nearly unuseable.
> 
> If I hit a page (RequestInfo example servlet) which doesn't do much
> and doesn't have turbine overhead the response it is much
> faster under tomcat 4...
> 
> BUT, if I run apache ab with count=1000 concurrent=50 the requests
> per second number I get for 3.2.1 is more than it is for tomcat 4
> hitting the RequestInfo servlet... This under green threads, if I
> run native threads in tomcat 3.2.1 it choakes on this test, but
> tomcat 4 doesn't
> 
> I also notice under low load that the page responses in the brower
> definately seem snappier under Tomcat 4..
> 
> Soo.. what is the significance of "native threads not supported" in
> SUN JDK1.2.2 for linux?
> 
> I've heard that there is only an advantage to running native-threads
> in an SMP situation, my results don't support that AT ALL..
> 
> There appear to be problems using native-threads in 3.2.1 but
> the same problems did not show up in tomcat 4... Does this mean
> I'm semi safe running native threads under tomcat 4?
> 
> This computer is much slower than the production dual PIII but
> it does give me some clue. This computer also only has 192megs ram
> and the JVM heap was not set to a max value, but it never went
> above about 18megs for total jvm size.
> 
> As far as load balancing which I do not have configured for a single
> tomcat instance on this machine... I can really load up
> one tomcat instance and fire of another one with the same webapp
> in it and response is instant.. so it is definately the JVM that
> is hurting and not the whole machine.  I figure it would
> take me 4 JVM's to get the same performance that I get with
> PHP on a similar page hit under the 20threads load.
> 


An update to this post...  Running IBM jdk1.3 -native threads
on against the same page, I get
1.3second average return for the exact same servelet page as tested
above only with 20 threads instead of the 10 above!!. 

This is actually better performance than I get out of a similar PHP
page which was much better than the previous jdk1.2.2/tomcat4
combination at handling the 10 threads, but PHP returned the
page in 2.6 seconds with 20 threads, so now my servlets are twice
as fast as my relatively similar php... Granted this is PHP3, still
I'm happy with the 1.3 seconds to return the servlet under that
load, like I said it is a turbine (old turbine) based servlet so
there is extra page layout overhead and such involved as well.

I didn't test the jdk1.3 under tomcat 3.2.1 but I've found tomcat4
is stable enough for what I'm doing right now.

Mime
View raw message