Results 1 to 4 of 4
  1. #1
    Junior Member
    Join Date
    Aug 2004
    Brussels, Belgium

    Websphere memory issue

    Hello all,

    I'm in an early stage of performance testing a quite complex application. I was checking out transaction response times, transaction per transaction, with a slow ramp up to 50 concurrent vusers. When I tried to maintain the load for an extensive period, the application suddenly stopped responding.

    I've been checking out several parameters:
    -Transaction response times rise linearly from the start of the scenario (ramp up 1 vuser per 10 sec)
    -The application server's CPU reaches 75% even at low load and doesn't scale up when more vusers are started.
    -Throughput is at its max after a few vusers have been added and slowly declines.

    The websphere monitor seems to pinpoint the problem:
    The "used memory" parameter fluctuates around a certain average for about an hour, the starts to rise gradually until the app no longer responds.

    Does anyone recognize this behaviour? Can anyone give a possible explanation so I can put our developers on the right track in solving this issue?

  2. #2

    Re: Websphere memory issue

    Well, there could be a lot of things going on here, but I would suspect an out of memory issue in the JVM. What do the websphere logs indicate? How does the garbage collection cycle look? What kind of think times do you have in between your requests? 75 virtual users with no think times, pounding away at the application as fast as they can will likely bring down any machine. Make sure you put realistic think times in between page/transaction requests.

  3. #3

    Re: Websphere memory issue

    Hi Cristax,

    your problem description sounds like a memory leak in the application. Try to check this out with appropriate tools.
    It is very strange, that the CPU load does not increase when adding more clients. At least the thread scheduler needs more CPU time when there are more clients. Do you use the Tivoli Performance Monitor? Check out whether the ORB Threads are always in use or the application has to wait for database or other connections. This might be the bottleneck and is of great interest for the developers.
    50 parallel users is not that much - depending on the hardware, of course.
    In my mind increasing the think time will hide the problem, but it is not solved. If the think times are really unrealistic, you would get Timeout Exceptions.

    If it is possible (and you're allowed to) send more information.

    Regards =)

  4. #4

    Re: Websphere memory issue

    -Transaction response times rise linearly from the start of the scenario (ramp up 1 vuser per 10 sec)

    I didn't notice this before. This looks like a database connection pool issue. We had a similar issue where the content management system only had 1 connection defined to the LDAP server, and this allowed only one transaction to go through at a time. Each successive transaction would wait for the one before it, so they would get longer and longer as the test went on. Check your database connection settings.



Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
BetaSoft Inc.
All times are GMT -8. The time now is 03:56 PM.

Copyright BetaSoft Inc.