• Apply the PatchQualityLevels to all the patches in the tree.
  • Audit the kgdb patches in the tree, to see what modifications were made specific to openmosix
    • Find out what version of kgdb this started out as

SetupENV Issues

  • Establish what the behavior of setupenv should be reguarding paths, relative and otherwise..

Userspace Issues

  • Document the test suite:
    • What it should do
      • Order of tests - first test should be signal, signal back, migrate, signal, signal back, migrate home, signal, signal back, die.
    • Document and commit changes required for execution under qemu
  • Document where the current code fails the test suite, and why, as far back as possible.
  • Compile and boot the current kernel (with the all the patches in series un-commented!) on all architectures.
  • Run the Linux test project on our kernels, and compare the results to un-patched versions. (make sure we didn't break anything.)

Kernelspace Issues

  • Examine patches 44 and 45, from openmosix-git.patch. See how some kernel functions take a task *, and others just assume current? It would be nice if someone would fix the linux kernel to have all the functions we're calling that assume current have an analogue that accepts a task *.


  • examine kdoc, find out how much of our docs can go in-tree.
  • Create a series of diagrams of subsystems similar to kerrighed's diagrams.




  • Raise awareness of the project.
  • Fix/Add? to our wikipedia article.

Public Services

  • Set up qemu hosts for public users to log into, so that they can run tests. web interface for the createvm suite?


  • examine how we handle extended register sets. MMX? SSE? SSE2? SSE3? XMM?
  • examine incorporation of tcp_nowait usage to reduce latencies.
  • examine incorporation of ECN usage cooperation with tcp_nowait.
  • examine both of the above with reguards to packets of size < 32 bytes in hpc/kcom.c around line 54.

Scheduler Issues

  • Add "opportunistic migration" option to code (similiar to casual 'job offer' model - advertising job, and then choosing best place to send it).
  • Integration of opportunistic migration routines with in-kernel processes monitoring (so i.e. tasks which are 'falling off' scheduler queue too often and are of lower priority will be given more migration oportunities.
  • opportunistic migration 'taxing' support (to allow shaping of load balancing using popular and commonsense 'economy' model - 'highly taxed' places are less attractive for migration)

Scheduler-Qemu-Kernel Interaction?

  • Possible sysfs control of oportunistic migration habits, taxation, and agressivenes?
  • Preemptive migration (migration of tasks before execution, i.e. migrating of native ppc code to real ppc machine from x86 home node)
  • Benefiting from virtualisation (instead of spawning guest process as userspace code, it can be spawned in virtual machine) - verify first if this will provide any benefit at all (generally low priority-cosmetics), though it could be nice if also in-kernel passing of processes to emulated virtual machines could be possible.
  • 'No-Execute Migration' -- better to migrate out than to swap out. support migrating a process to a node not designed to run it, as a temporary store.

VMA Issues

  • Verify that we have implemented demand-paging for VMA pages from the home process. haven't seen the code for this yet...
Last modified 8 years ago Last modified on Oct 12, 2013, 7:12:09 AM