Well 
I had the impression that the pure XML-handling was a killer on the solaris box.
 
As with JSF: most processing power is used in the jsp-rendering. I don't know 
whether jsp-precompiling might work or be usefull, I never tried.
I the meantime I moved to facelets, which accelerates everything... even though
it uses a SayCompiler, but it seems that Compiler is blindingly fast...
 
regards
Alexander

________________________________

From: Martin Denham [mailto:[EMAIL PROTECTED] 
Sent: Tuesday, July 24, 2007 5:46 PM
To: MyFaces Discussion
Subject: Re: 4 second page response time


Hi Jesse,

I thought the ContextLoaderListener would just be called once on startup.  Does 
it do something on every request?

Martin


On 24/07/07, Jesse Alexander (KSFD 121) <[EMAIL PROTECTED]> wrote: 

        I remember that long time ago I did a comparision for the same reason...
        and we found out, that the xml-processing on the solaris box was WAY 
slower than on the
        Win-Box.
        We never really found out why, though...
         
        regards
        Alexander

________________________________

        From: Martin Denham [mailto:[EMAIL PROTECTED] 
        Sent: Tuesday, July 24, 2007 4:48 PM
        To: MyFaces Discussion
        Subject: Re: 4 second page response time
        
        
        
        Thanks for the tips but our sys admins weren't keen on doing a kill and 
because I only get performance problems on the central Solaris server it was 
tricky to follow your advice.  I checked for missing tld/xsd warnings and we 
aren't getting any even though, as you guessed, the Solaris server does not 
have internet access. 
        
        However, I have managed to find another of our jsf applications which 
did not have the 4/8 second page response delay and so I slowly migrated this 
to be more like the troublesome application.
        
        The main problem occurs when I include 
            <listener>
                <listener-class>
                    org.springframework.web.context.ContextLoaderListener
                </listener-class>
            </listener>
        in web.xml.  Yes, the problem also occurs if I use ContextLoaderServlet 
too. 
        
        After including ContextLoaderListener performance deteriorates 
considerably from 2 second response to more than 4 even if I don't load any 
spring contexts.
        
        Has anybody any idea why ContextLoaderListener slows down my 
application running on Weblogic 8.1 on Solaris?
        
        Many thanks.
        
        Martin
        
        
        On 23/07/07, David Delbecq <[EMAIL PROTECTED]> wrote: 

                En l'instant précis du 20/07/07 15:02, Martin Denham 
s'exprimait en ces
                termes: 
                > I have had a performance issue with both the JSF applications 
I have
                > written.
                >
                > On my windows xp development pc responses are instant.  
However when
                > deployed to a Sun Ultra 80 Solaris machine every page takes 4 
seconds 
                > and if I add a redirect the response time increases to 7 
seconds.
                > Another application on the same Solaris machine, but written 
using
                > Struts has instant page response times.
                >
                > Is a simple page response time of 4 seconds expected when 
using JSF? 
                > I have tried all sorts of tweaks during the past year but the 
response
                > time is unaffected.
                Simple answere: no. I will have to profile your application to 
find out
                where your CPU bottleneck (if it's a CPU bottleneck) is, or 
where your 
                network bottleneck is. Because JSF uses value binding which can 
do lots
                of things, any badly written/badly used bean can be at cause 
(like a
                bean loading 50.000 items for a database at each request).
                Simple suggestion: 
                when you load a JSF page, go in a console to your solaris 
station and
                run a kill -3 <JVMpid>, this will dump to the jvm's stdout a 
stacktrace
                of all running threads. From there you could see where the code 
is 
                waiting / busy.
                
                could it be some xml parser uses a xsd/dtd which is not 
available. If
                production server is firewalled, maybe the server is just 
trying to
                download the schema/dtd and finishes on a timeout of approx 4 
seconds? 
                >
                > I am using Myfaces & tomahawk 1.1.5,  Weblogic 8.1sp4, 
Facelets
                > 1.1.12.  One application uses Oracle ADF and the other 
Ajax4Jsf/Richfaces.
                >
                > Thanks in advance for any pointers.
                >
                > Martin
                >
                >
                
                
                --
                http://www.noooxml.org/
                
                



Reply via email to