Loading ...
Sorry, an error occurred while loading the content.

Strange CPU utilization behavior

Expand Messages
  • David Gewirtz
    I m experiencing some weird strangeness that highlights Dave s request for a profiler. I m posting this because I m not exactly sure where to look, and I can t
    Message 1 of 3 , Feb 2, 2005
    • 0 Attachment
      I'm experiencing some weird strangeness that highlights Dave's
      request for a profiler. I'm posting this because I'm not exactly
      sure where to look, and I can't tell if I should be digging into the
      kernel or it's a coding issue.

      For some number of months, Frontier would, every few days, peg at
      100% CPU utilization. When it pegged, Frontier's responsiveness
      slowed down to a crawl. And, once it went to 100%, it would stay
      there until a restart. But, if I restarted Frontier, the pegged
      utilization went away for a few days. This went on until about Jan
      5.

      Then, something new and exciting happened. Frontier no longer pegged
      the CPU, but instead the PF usage climbed regularly, often getting
      to levels around 600 or 800MB.

      This weekend, things changed again. I'm now seeing almost constant
      100% utilization, although once in a while, it drops back down to
      1%. I'm also getting reasonable responsivness.

      I suspect someone might be crawling the site, or some script is
      getting triggered. But I'm not finding any good ways to diagnose
      this. I'd love to know what's going on and fix the problem.

      So, this is both a supporting request for profiling as well as a
      plea for help. Any ideas?

      -- David
    • Karsten Wolf
      Hi, Am 02.02.2005 um 18:47 schrieb David Gewirtz: ... Haven t seen this. But I run Frontier not as a server nor the Manila framework. ... I m a bit concerned
      Message 2 of 3 , Feb 4, 2005
      • 0 Attachment
        Hi,

        Am 02.02.2005 um 18:47 schrieb David Gewirtz:
        ...
        > For some number of months, Frontier would, every few days, peg at
        > 100% CPU utilization. When it pegged, Frontier's responsiveness
        > slowed down to a crawl. And, once it went to 100%, it would stay
        > there until a restart. But, if I restarted Frontier, the pegged
        > utilization went away for a few days. This went on until about Jan
        > 5.
        Haven't seen this. But I run Frontier not as a server nor the Manila
        framework.
        >
        > Then, something new and exciting happened. Frontier no longer pegged
        > the CPU, but instead the PF usage climbed regularly, often getting
        > to levels around 600 or 800MB.
        I'm a bit concerned that on OSX real memory usage, as displayed by top
        goes up to 160MB just for traversing a 16MB Frontier.root. This could
        be related. I regard it more as an annoyance, not a real bug...

        > This weekend, things changed again. I'm now seeing almost constant
        > 100% utilization, although once in a while, it drops back down to
        > 1%. I'm also getting reasonable responsivness.
        Have you changed the kernel?
        >
        > I suspect someone might be crawling the site, or some script is
        > getting triggered. But I'm not finding any good ways to diagnose
        > this. I'd love to know what's going on and fix the problem.
        >
        > So, this is both a supporting request for profiling as well as a
        > plea for help. Any ideas?

        Only an idea: I always wanted to use Sherlock (Ed Reams debugging tool,
        not the apple search tool
        http://sourceforge.net/project/showfiles.php?group_id=3458
        file:sherlock002.zip
        http://personalpages.tds.net/~edream/ )
        a dynamic profiling & debugging tool. But since it must be integrated
        with every function (frontier has ca. 6000 functions) I requested (&
        got) the missing integration tools from the author. Haven't tried to
        compile them yet (old mac sources) and will not have the time for in
        the next two weeks.

        Sherlocks profiling tools can be switched selectively at runtime. This
        could be very useful if the "misbehaviour" starts two month after
        launch.

        But... this is a complex package. There must be some evaluation if it's
        worth the trouble. It includes a tool for automatic insertion of the
        macros (SPP) but I wouldn't expect it to run on the first try.

        I will have a serious try at this but not now. If s.o. else wants to
        take the first step (or trying parallel) I can email the complete
        package.

        regards
        karsten
      • Terry Teague
        ... While I can t help with installations running on Windows, I did suggest earlier that any Mac OS X developers in this situation should use something like
        Message 3 of 3 , Feb 5, 2005
        • 0 Attachment
          --- In frontierkernel@yahoogroups.com, Karsten Wolf <karstenwo@w...> wrote:
          > Am 02.02.2005 um 18:47 schrieb David Gewirtz:
          > ...
          > > For some number of months, Frontier would, every few days, peg at
          > > 100% CPU utilization. When it pegged, Frontier's responsiveness
          > > slowed down to a crawl. And, once it went to 100%, it would stay
          > > there until a restart. But, if I restarted Frontier, the pegged
          > > utilization went away for a few days. This went on until about Jan
          > > 5.

          > > Then, something new and exciting happened. Frontier no longer pegged
          > > the CPU, but instead the PF usage climbed regularly, often getting
          > > to levels around 600 or 800MB.

          > > This weekend, things changed again. I'm now seeing almost constant
          > > 100% utilization, although once in a while, it drops back down to
          > > 1%. I'm also getting reasonable responsivness.

          > > I suspect someone might be crawling the site, or some script is
          > > getting triggered. But I'm not finding any good ways to diagnose
          > > this. I'd love to know what's going on and fix the problem.
          > >
          > > So, this is both a supporting request for profiling as well as a
          > > plea for help. Any ideas?


          While I can't help with installations running on Windows, I did suggest earlier that any Mac
          OS X developers in this situation should use something like Shark or similiar tools. I
          haven't had time to try Shark on Frontier yet, but I know others have.


          > Only an idea: I always wanted to use Sherlock (Ed Reams debugging tool,
          > not the apple search tool
          > http://sourceforge.net/project/showfiles.php?group_id=3458
          > file:sherlock002.zip
          > http://personalpages.tds.net/~edream/ )
          > a dynamic profiling & debugging tool. But since it must be integrated
          > with every function (frontier has ca. 6000 functions) I requested (&
          > got) the missing integration tools from the author. Haven't tried to
          > compile them yet (old mac sources) and will not have the time for in
          > the next two weeks.


          Cool. I was aware of Ed's other contributions, mainly LEO. He seems like quite a clever (and
          helpful) guy.


          > Sherlocks profiling tools can be switched selectively at runtime. This
          > could be very useful if the "misbehaviour" starts two month after
          > launch.
          >
          > But... this is a complex package. There must be some evaluation if it's
          > worth the trouble. It includes a tool for automatic insertion of the
          > macros (SPP) but I wouldn't expect it to run on the first try.


          When I was looking for some modern (preferably open source) instrumentation code to use
          for profiling/tracing code that was running in an environment where the normal Mac OS X
          libraries and debugging tools weren't available (I was trying to track down a hard to find
          bug), I found and used CodeWorker (with some success) :

          <http://codeworker.free.fr/>


          > I will have a serious try at this but not now. If s.o. else wants to
          > take the first step (or trying parallel) I can email the complete
          > package.


          I don't really have time to help at the moment, but I am interested in anything you might
          learn from this.


          > regards
          > karsten


          Regards, Terry
        Your message has been successfully submitted and would be delivered to recipients shortly.