You are viewing a plain text version of this content. The canonical link for it is here.
Posted to general@jakarta.apache.org by Vasile GABURICI <ga...@cs.pub.ro> on 2000/06/02 19:24:49 UTC
Re: [AOLSERVER] ADP vs JSP performance
On Fri, 2 Jun 2000, Preston L. Bannister wrote:
> From: Vasile GABURICI [mailto:gaburici@cs.pub.ro]
> > I did a bit of testing regarding the performance of ADP and JSP
> > on AOLserver. The results are quite surprising. Using the latest generation
> > in-process JVM, a JSP is twice faster than its ADP counterpart. Details on:
> >
> > http://www.ss.pub.ro/~gaburici/nstomcat/perf/
>
> Great stuff! :) I was curious about exactly this question.
>
> ... though I suspect the following statement is inaccurate:
>
> " If this test were conducted a few months ago, before the
> JDK 1.3 was upon us, the ADP would have won. Now with the
> new breed of JVMs, the crown goes to JSP, no contest. "
>
> Both the IBM 1.1.8 and 1.3 JDKs benchmark at radically high performance
> levels compared to all the JDK's previously available on Linux. Credit
> is due to IBM for bringing high performance Java to Linux.
I presume that you are aware that IBM JDK 1.1.8 cannot be
currently used properly with nstomcat if the JIT is enabled.
To answer your concern, I also did some testing with stand-alone
Tomcat using the pool TCP connector. The speed is almost the same that JNI
connector provides. Beware that the JNI connector is not yet optimized,
while the TCP pool was tuned by Gal Shachor.
Anyway, on stand-alone tests, IBM JDK 1.1.8 does about 30 req/sec,
while the other two JDKs provide results close to those obtained in JNI
mode. So IBM JDK 1.1.8 is a tad faster than Blackdown 1.2.2 (17 req/sec),
but not anywhere near the 70+ req/sec that 1.3.0 pulls.
> So the difference in Java performance is most likely not due to Java
> 1.2 versus Java 1.3, rather it's Blackdown versus IBM's implementation.
No entirely so, but it is true that IBM makes some great JDKs.
> --
> Preston L. Bannister
> preston@home.com
> http://members.home.com/preston
> pbannister via Yahoo! Messenger
>
> This is the AOLSERVER list, an open forum from http://aolserver.com/
> To sign-off, send "SIGNOFF AOLSERVER" to listserv@listserv.aol.com
> Subscribe to AOLSERVER-ANNOUNCE for announcements and news.
>
Re: [AOLSERVER] ADP vs JSP performance
Posted by Vasile GABURICI <ga...@cs.pub.ro>.
Actually Costin wrote me some of the tips bellow before I did the
test, so I was aware that further optimizations were possible on the JSP
side. I didn't use them in the tests because I didn't know how to do
similar optimizations for ADP. So, to make a fair test, I wrote the JSP
and ADP as an average user would do.
Even so, I'm already getting flamed on the AOLserver list for
using the magic weapon 'for' which is said to be much slower on TCL :-0
With Costin's permission, I'd like to forward this message to the
AOLserver list, so people there might get that I didn't try to bash ADP...
On Fri, 2 Jun 2000, Costin Manolache wrote:
>
> I would like to add few things to this topic : it seems the tests
> are done with "default" tomcat configuration.
>
> By using jspc to precompile the pages ( and declare them in
> web.xml - avoiding JspServlet ) and with few changes in the
> jsp implementation that will be part of 3.2, you can easily
> double the performance.
>
> On my Celeron/366 I get around 50 req/sec with a similar test,
> but around 120 req/sec by reusing the Jsp writer.
> ( unfortunately this has a bug and fails some tests, but it will
> probably be fixed for release ).
>
> We'll get another big improvement by fixing the char->byte
> conversion.
>
> I don't know if all this will get into 3.2 or we'll save some
> for 3.3, but expect some big improvements in speed - and this is
> just the start !
>
> With "experimental" version of tomcat ( i.e. few hacks to bypass
> the conversion and few hacks to avoid some expensive code - that
> we still have to fix ) - I can get close to 80% of the speed of Apache 2.0
> serving static files, and in some cases Apache2.0 + Jni + tomcat ( or even
> tomcat standalone ) is faster than Apache 1.3 with static files.
> Again, this is not something you can use in production, there is
> still a lot of work.
>
> Costin
>
> ( that without even doing any "special" optimizations ! )
>
>
>
>
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: general-unsubscribe@jakarta.apache.org
> For additional commands, e-mail: general-help@jakarta.apache.org
>
>
Re: [AOLSERVER] ADP vs JSP performance
Posted by Costin Manolache <co...@eng.sun.com>.
I would like to add few things to this topic : it seems the tests
are done with "default" tomcat configuration.
By using jspc to precompile the pages ( and declare them in
web.xml - avoiding JspServlet ) and with few changes in the
jsp implementation that will be part of 3.2, you can easily
double the performance.
On my Celeron/366 I get around 50 req/sec with a similar test,
but around 120 req/sec by reusing the Jsp writer.
( unfortunately this has a bug and fails some tests, but it will
probably be fixed for release ).
We'll get another big improvement by fixing the char->byte
conversion.
I don't know if all this will get into 3.2 or we'll save some
for 3.3, but expect some big improvements in speed - and this is
just the start !
With "experimental" version of tomcat ( i.e. few hacks to bypass
the conversion and few hacks to avoid some expensive code - that
we still have to fix ) - I can get close to 80% of the speed of Apache 2.0
serving static files, and in some cases Apache2.0 + Jni + tomcat ( or even
tomcat standalone ) is faster than Apache 1.3 with static files.
Again, this is not something you can use in production, there is
still a lot of work.
Costin
( that without even doing any "special" optimizations ! )