S5R4 on Linux x64 errors out

Forums › Cruncher's Corner

Metod, S56RKO
Metod, S56RKO
Joined: 11 Feb 05
Posts: 135
Credit: 834045947
RAC: 83163
12 Aug 2008 13:27:50 UTC
Topic 193848
(moderation:
    )

    I have an Opteron 2218 (dual core) Debian Lenny which seemingly developed dislike for Einstein. Something like half a year ago it wasn't able to complete a S5R3 WU without problem. With S5R4 I gave it another go and indeed it was successful on almost 20 WUs. Today it choked on another one, catching SIG8: http://einsteinathome.org/task/103577395. It has some backtrace, can some developer take a look?

    This machine quite nicely runs some other projects (SaH - both seti_enhanced (the KWSN variant) and astropulse - and Rosetta@Home - all apps).

    Metod ...

    Bikeman (Heinz-Bernd Eggenstein)
    Bikeman (Heinz-...
    Moderator
    Joined: 28 Aug 06
    Posts: 3522
    Credit: 832623407
    RAC: 1208949

    S5R4 on Linux x64 errors out

    12 Aug 2008 14:14:08 UTC
    Message 84289
    (moderation:
      )

      There is a [url=http://www.nabble.com/CONFIG_PREEMPT-causes-corruption-of-application's-FPU-stack-td17293854.html]known bug[/url] in the Linux kernel that was only recently fixed, it is responsible for quite a few "signal 8" errors for Einstein@Home (actually a E@H user played an important role in identifying the bug and helped to analyze it. We may not have found GWs yet, but at least a Linux kernel bug ;-) ).

      The bug will be visible when the kernel is compiled with the CONFIG_PREEMPT=y setting (often set in "real-time" kernel variants).

      Other potential reasons for "sig 8" problems include overclocking, undervolting, poor cooling .... but I'd go after the kernel bug possibility first. Updating the kernel to a recent one could cure this if you are lucky.

      Bikeman

      Metod, S56RKO
      Metod, S56RKO
      Joined: 11 Feb 05
      Posts: 135
      Credit: 834045947
      RAC: 83163

      Thanks for the pointers,

      12 Aug 2008 19:17:33 UTC
      Message 84290 in response to message 84289
      (moderation:
        )

        Thanks for the pointers, Bikeman!

        It sounds surrealisticaly familiar ... custom-built kernel with CONFIG_PREEMPT=y etc. in my case. I can hardly believe that other reasons you've enumerated can be true in my case as this is a nice HP workstation designed to take two CPUs while only one is installed ATM. No overclocking etc. Which reminds me that I need to install another CPU and a couple of gigs of RAM before AMD becomes extinct :-)

        Metod ...

        Metod, S56RKO
        Metod, S56RKO
        Joined: 11 Feb 05
        Posts: 135
        Credit: 834045947
        RAC: 83163

        I've installed new kernel on

        14 Aug 2008 12:24:43 UTC
        Message 84291 in response to message 84290
        (moderation:
          )

          I've installed new kernel on my machine:

          $ uname -a
Linux ural 2.6.26.2 #2 SMP Thu Aug 14 08:42:29 CEST 2008 x86_64 GNU/Linux

          I think that this one shouldn't experience the same bug anymore. I've also changed the configuration of preemption to a slightly less aggressive one:

          # CONFIG_PREEMPT_NONE is not set
CONFIG_PREEMPT_VOLUNTARY=y
# CONFIG_PREEMPT is not set

          Hopefully this marks this thread as closed/solved.

          Metod ...

          Comment viewing options

          Select your preferred way to display the comments and click "Save settings" to activate your changes.

          Forums › Cruncher's Corner