#7 Remote nodes very slow

beta_pre-1.0
closed-accepted
9
2006-03-02
2006-03-01
No

When running the Demo in the JPPF distribution, adding
a 2nd, remote, node does not significantly speed-up
the execution. I some cases, it even makes it a lot slower.

This problem has been reported initially in the help
forum, in this this thread:
http://sourceforge.net/forum/forum.php?thread_id=1450614&forum_id=458549

This issue is assigned the highest priority.

Discussion

  • Laurent Cohen

    Laurent Cohen - 2006-03-02
    • status: open --> closed-accepted
     
  • Laurent Cohen

    Laurent Cohen - 2006-03-02

    Logged In: YES
    user_id=1252190

    The issue is effectively that the ratio TCP transport time /
    task execution time is too high.

    This is especially visible when the number of nodes is small
    (like 2 or 3 for instance), or the network connection
    quality is low.

    In this type of situation, you'll always end up with one or
    more nodes spending more time in network I/O than in actully
    tasks execution, which explains the low CPU utilization rate.

    In the current JPPF demo, the tasks are very short (15 ms to
    50 ms in average), so this problem is effectively visible.
    Also each node receives an entire 300*300 matrix of double
    values in compressed format, which amounts to around 50-60
    KB sent over to the node.

    I will add another example demo, in a (near) future, with
    tasks that last much longer (in the order of 1 second),
    which should a much more signifiant performance gain.

     

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks