Hey - thanks guys - I should have posted a followup myself
(which I guess this is :-) and thanks for the JProbe suggestion
Andy, I'll take a look. And yes - using 1.3.
Sean, I haven't seen the blank page thing?? I'll look but I
haven't seen any indications in the logs of exceptions.
I reworked the overall configurations while getting struts
up for play. I also cut the logging down from info to warn.
When I re-ran the tests there was remARkable improvement in
performance...
The new times for 1,000 requests and 50 concurrent users came in as...
static html 2.5 secs 392.3 req/sec 86% improvement
servlet snoop 4.1 secs 244.1 req/sec 86% improvement
nws version 21.9 secs 45.6 req/sec 41% improvement
jsp snoop 5.9 secs 168.3 req/sec 95% improvement
... compared to the old times of...
static html 17.6 secs 56.9 req/sec http://216.62.183.108/snoopResults.htm
servlet snoop 28.3 secs 35.3 req/sec http://216.62.183.108/servlet/snoop/
nws version 37.1 secs 26.9 req/sec http://216.62.183.108/snoopNWSstyle.htm
jsp snoop 130.3 secs 7.7 req/sec http://216.62.183.108/jsp/snoopBare.jsp
I think a lot of this improvement actually has to do with
lower local network traffic which seems to impact the ab
testing in general. But, for sure, I think the jsp was
improved by the reconfiguration as well. (NWS doesn't take
advantage of the multiprocessor platform so I think that's
why it improved less than the others).
THis example is obviously a VERY simple page - very little actual
processing, but I wanted to get a sore of baseline for use when
looking at more realistic use.
Anyway - thanks for the follow-ups - and I'll look forward to
even more improvements as tomcat et all mature a bit.
Andy Nuss wrote:
>
> Use JProbe to verify that 95% of the time is spent in
> out.print("your static text"), then take it from there.
>
> Also, make sure you are using JDK 1.3, which is much
> faster than JDK 1.2 for lots of things.
>
> (Be patient, performance tuning for JSPs will come soon.)
>
> -----Original Message-----
> From: Sean Blaes
> To: '[EMAIL PROTECTED]'
> Sent: 3/9/2001 1:17 PM
> Subject: RE: load test performance comparisons - again...
>
> Have you tuned your configuration and startup files?
>
> Also, when hitting your site, I've found that it often comes up with a
> blank
> page. Perhaps there is an error in your code which is throwing
> exceptions
> which could slow your application down...
>
> -----Original Message-----
> From: Horace A. Vallas, Jr. [mailto:[EMAIL PROTECTED]]
> Sent: Tuesday, March 06, 2001 11:32 PM
> To: [EMAIL PROTECTED]
> Subject: load test performance comparisons - again...
>
> Hi all - I know - I know - beat that horse to death!! sorry but...
>
> I'm still relatively new to tomcat and playing with maybe
> migrating a couple of fairly large and active sites to jsp/servlets.
>
> running 1.3.12 with 3.2.1 and jdk2-13 on RH7/2.2.16-22smp on a
> 2-processor i686 with 256MB (I think) - it's not a production server
> but should be sufficient for development and some thumbnail testing
> I guess.
>
> I used apacheBench to run a quick test for loads of 50 concurrent
> users and 1000 requests - not an extremely heavy load -
> I compared a small version of snoop.jsp with a static html copy
> of the jsp's rendered results and a similar report generated using
> the neowebscript apache mod (like a tcl php)
>
> As expected the static html was fastest, followed by the nws version and
> the
>
> jsp, respectively - but the difference in the jsp was startlingly
> large...
>
> static html 17.6 secs 56.9 req/sec
> http://216.62.183.108/snoopResults.htm
> nws version 37.1 secs 26.9 req/sec
> http://216.62.183.108/snoopNWSstyle.htm
> jsp snoop 130.3 secs 7.7 req/sec
> http://216.62.183.108/jsp/snoopBare.jsp
>
> Running the same tests with more realistic number of requests given
> current
> site loads (-n 400 -c 50) yielded similar differences...
>
> static html 8.2 secs 48.9 req/sec
> http://216.62.183.108/snoopResults.htm
> nws version 16.5 secs 24.2 req/sec
> http://216.62.183.108/snoopNWSstyle.htm
> jsp snoop 54.8 secs 7.3 req/sec
> http://216.62.183.108/jsp/snoopBare.jsp
>
> I'm floored and discouraged by the speed (slowness) of the tomcat stuff?
> and we're not even talking about any database use etc. in this little
> test.
>
> Can anyone point me to some basic flaw in this as a comparison?
> Maybe ab is causing each jsp access to incur compile time or something?
>
> I was really really looking forward to being able to migrate from
> tcl to java but this sort of comparative performance is going to
> make it a VERY difficult sell.
>
--
Wishing you an "OOBA OOBA" 21st Century!!
Horace ...once known as "Kicker" :-)
================================================================
Horace Vallas hav.Software http://www.hav.com/
P.O. Box 354 [EMAIL PROTECTED]
Richmond, Tx. 77406-0354 voice: 281-341-5035
USA fax: 281-341-5087
Thawte Web Of Trust Notary in SW Houston, Tx.
http://www.hav.com/?content=/thawteWOTnotary.htm
================================================================
...drop by and chat if I'm online http://www.hav.com/chat/
=== === === === === === === === === === ====
What is a Vet? ... He is the barroom loudmouth, dumber than five
wooden planks, whose overgrown frat-boy behavior is outweighed a
hundred times in the cosmic scales by four hours of exquisite
bravery near the 38th parallel. ... - Unknown
================================================================
S/MIME Cryptographic Signature
