4 Replies Latest reply on Oct 6, 2010 6:56 PM by yahawari

    stress test: results are far from satisfying

    walidkarray
      We are stress testing our Jboss Seam web application but we discovered that's results are far from satisfying.
      As the first test case we supposed 01 Thread/User sending 20000 HTTP GET requests successively on an URL that returns a blank page "http://10.10.0.183:8080/project/empty.seam" knowing that there is no DB call neither method invocation, as it finishes; statistics show an average of 07.4 Requests/s!!! while the average we expected was about 100 Requests/s.

      Note: For the stress test simulation tool we have used the latest version of Apache JMeter.

      * Hardware environment:
      --------------------------
      - Server HP ProLiant DL320 G6,
      - CPU: Intel(R) Xeon(R) CPU 5520 @ 2.27GHz x 8 (8 Core) [64 Bits],
      - Memory: 12 GB,
      - 500 GiB HDD.

      * Software environment:
      --------------------------
      - Operating system: Redhat Enterprise Edition 5 (64 Bits)
      - JVM: OpenJDK 64-Bit Server VM (build 1.6.0-b09, mixed mode)
      - JBoss EAP 5.0.0 (Intgrate JBoss ESB 4.8)
      - JBoss Seam 2.2
        • 1. Re: stress test: results are far from satisfying
          mmaia

          Have you monitored cpu usage and idle time in both sides? I mean in the server side and also in the client machine running JMeter as I have already been in a scenario where client was overloaded by jmeter pretty much before the server I was trying to stress. If you notice this is the case you will need to use more client machines to run your tests.


          regards.

          • 2. Re: stress test: results are far from satisfying
            walidkarray
            Thanks Marcos for your fast reply,

            We have optimized throughput average to 20/s by removing three unused components:("drools, jbpm and mail") in "components.xml" but this average still inadequate for a "non business processing" call.

            For the monitoring we did it by installing "Munin" both in client and server side to graph performances during the load test. Results are detailed below:

            - Server side:
            Load average=0.75
            cpu= 10%
            mem= 60% used
            swap= 0%

            - Client side:
            load average=0.60
            cpu= 16%
            mem= 40% used
            swap= 0%

            => You can easily note that both client and server are lightly loaded !

            I have already made another different load test "http://seamframework.org/Community/AnWeRelyOnJBossSeamToSupportHeavyLoadCalls" and I discovered that's the long-time response comes from seam core components.

            My best regards.
            • 3. Re: stress test: results are far from satisfying
              lvdberg

              Hi,


              see the answer in your other thread. You can't compare if you're not using a realistic scenario. For every Seam enabled tiny page, you indeed have a lot of overhead. But his is not very realistic. We seldom have applications containing only empty pages.



              Leo

              • 4. Re: stress test: results are far from satisfying
                yahawari

                if i am not mistaken, seam creates a transaction that wraps every request. to have an exact similar scenario, your two pages (jsf and jsf/seam)should get a couple of db records and display them.


                u need to run seam in production mode as well (debug off).