4 Replies Latest reply on May 4, 2004 10:56 AM by Remy Maucherat

    Bechmark , tomcat5 alone VS jboss3.2.3(with tomcat5)

    Tomer ben david Novice

      Hi

      I performed some benchmarks to my application first using tomcat5 only and then using a jboss3.2.3 with tomcat5 (I tested only the web/servlet/jsp part of my application)

      12 Hours
      10 Concurrent users
      Both tests were performed with the default configuraiton of tomcat5 and jboss3.2.3
      With webload
      (no need to mention a clean machine before each test, and I repeated these tests and they came out with approximately the same results each time...)

      Results

      1. Tomcat5

      Avg Hits per second : 36
      Avg Throughput : 435153

      4. JBoss3.2.3+tomcat5

      Avg Hits per second : 15.978
      Throughput : 212232


      Can Anyone explain to me how come I received such different results in the second configuration?! it is the same tomcat5 right?! Can anyone point to some configuration updates / tuning I should perform to jboss so that it will reach tomcat5 stand alone results?!

      Thanks...

        • 1. Re: Bechmark , tomcat5 alone VS jboss3.2.3(with tomcat5)
          Remy Maucherat Novice

          Try 3.2.4 RC 1. Also make sure the Jasper settings are right (= it's not running in dev mode), as this can have a very big impact.

          In 3.2.3, there were obvious problems with the default configuration:
          - inappropriate logging levels
          - access logging

          I believe the Tomcat 4.1 configuration has similar performance issues by default. This can be addressed by tweaking configuration files, of course.

          I recently profiled and optimized the defalt configuration, and it is now almost identical to Tomcat 5 standalone (the difference, about 5%, being some extra J2EE related features you can disable).

          • 2. Re: Bechmark , tomcat5 alone VS jboss3.2.3(with tomcat5)
            Tomer ben david Novice

            performed the benchmark again:

            * I used JBOSS 3.2.4RC1 instead of JBOSS 3.2.3
            * Updated the jasper settings (JspServlet) to

            <init-param>
            <param-name>logVerbosityLevel</param-name>
            <param-value>FATAL</param-value>
            </init-param>
            <init-param>
            <param-name>classdebuginfo</param-name>
            <param-value>false</param-value>
            </init-param>
            <init-param>
            <param-name>development</param-name>
            <param-value>false</param-value>
            </init-param>
            <init-param>
            <param-name>reloading</param-name>
            <param-value>false</param-value>
            </init-param>

            The results are a bit better

            Avg Hits per second : 19.5 (improvement from 15.978)
            Avg Throughput : 259398 (improvement from 212232)

            But its nothing to compare with the tomcat5 stand alone results (Avg Hits per second : 36 , Avg Throughput : 435153)

            Since these are the results I will try to follow the whole unofficial guide for better performance and notify of the results.

            If there are any strong recommendations for other issues for me to tune I would be thankfull as there are yet very large differences from a stand alone tomact5 to jboss324rc1 + tomcat5...

            • 3. Re: Bechmark , tomcat5 alone VS jboss3.2.3(with tomcat5)
              Remy Maucherat Novice

              I did compare with a dumb static file, which I thought was a good enough test. I will look at JSP exclusively then, but I recommend testing with a simple static resource.

              You have to make sure there's no extra stuff in the JBoss configuration, such as access logging.

              • 4. Re: Bechmark , tomcat5 alone VS jboss3.2.3(with tomcat5)
                Remy Maucherat Novice

                I did test again, using the precompiled jsp-examples webapp from TC 5. Good thing I did since I ran into a performance issue (now fixed), but I think it was not present in 3.2.4 RC 1.

                Since you're testing JSPs, you should check that you're not actually testing session creation speed (it could be slower if JBoss clustering is enabled, for example).
                Also be careful that a lot of things have a performance cost. One example is the use of the "SubjectAttributeName" (but this should have been ok with 3.2.4 RC 1, as this wasn't implemented yet).

                Let me know if you solve your performance issues, or if you can help me reproduce the problems.