Announcement

Collapse
No announcement yet.

Multicore tests don't actually use multicore, where to submit patches?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Multicore tests don't actually use multicore, where to submit patches?

    I noticed that it appears _none_ of the tests that the multicore suite uses... actually use more than one core. Most simply need modifications to their run scripts to pass a command line option using $NUM_CPU_CORES. I'm sort of figuring out that the tests are hosted on OBM, but I can't find anywhere to ask over there, and there's no bug tracker or anything that I can find either. I don't mind patching these up but I can't find info on where to submit anything to a maintainer. Who is 'in charge' of maintaining the tests?

    Thanks,
    Jim

  • #2
    Okay, on second glance, several of them are running parallelized already, some simply aren't multithreaded software, and some just need some love. Here's the list of ones that aren't, and link to fixed install.sh's for the ones I could.


    pts/openssl -- as far as I can tell, this simply isn't parallelized. Should be removed from multicore suite?
    pts/gcrypt -- won't compile on my system
    pts/john-the-ripper -- fixed
    pts/himeno -- the algorithm could be parallelized, but it's not, not going to mess with it. Should be removed?
    pts/bullet -- got it running multithread, but it seems to be broken.
    pts/povray -- rendering itself isn't multithreaded, although scripts could be done to use multiple processes. Should be removed?
    pts/crafty -- tried to fix, but took WAY longer to run... Not sure what may have broke.
    pts/mafft -- fixed
    pts/minion -- not parallelized?? Should be removed?

    Here's a link to the patched install.sh's for john-the-ripper and mafft. Benchmark differences:

    http://openbenchmarking.org/result/1201176-AR-MULTICORE59


    I was using my computer otherwise during both tests, so the 'After' had a lot more deviation than it normally would, but still, it's pretty obvious they were only using 1 core beforehand.

    Jim

    Comment


    • #3
      Normaly you can use openssl speed rsa -multi $No_CPU_CORES. Is it possible to add this?

      Comment


      • #4
        My question is rather similar. I noticed that some tests are quite sub-optimal, but there is no place to submit fixes for them? I could only find the git for PTS core, not for the tests themself.

        Comment


        • #5
          Originally posted by keitai View Post
          My question is rather similar. I noticed that some tests are quite sub-optimal, but there is no place to submit fixes for them? I could only find the git for PTS core, not for the tests themself.
          The test profiles themselves are downloaded automatically from OpenBenchmarking.org when needed and placed inside ~/.phoronix-test-suite/test-profiles/ For now the easiest way to submit any changes to test profiles would be to the mailing list or email michael at phoronix.com.
          Michael Larabel
          https://www.michaellarabel.com/

          Comment


          • #6
            Originally posted by jimduchek View Post
            Here's a link to the patched install.sh's for john-the-ripper and mafft. Benchmark differences:

            http://openbenchmarking.org/result/1...AR-MULTICORE59

            Pulled them in for MAFFT and JTR on OpenBenchmarking.org now, thanks. Apologies for the late response, somehow didn't notice the thread previously. (Anyone is always welcome to ping michael at phoronix.com).
            Michael Larabel
            https://www.michaellarabel.com/

            Comment

            Working...
            X