ABOUT

A test matrix replacement

Why?
  Current approach is to manually pick some tests out of a matrix. This is
  done more or less random with human interaction and may lead to ignoring
  tests which failed in the past or running some images more often than
  others. Also, the configuration variance of the matrix is limited for a
  single image.

How?
  There are two approaches needed:
    For manual testing it's better to do the scheduling per host, as each
    tester does have his `own' boxes. This will be done first.
    For automatic testing through Tapper the scheduling should be done per
    testee (xen-unstable, xen-stable-testing, kvm). This should require
    minor additions only.

  Scheduling:
    - OS vendors round robin
    - Image / test combinations for each vendor somewhat round robin,
      but depending on system configuration, resources, and finally
      randomness. Goal is to run every possible combination once per
      cycle.
    - Guest memory and VCPU configuration depending on image type,
      optimal usage of system resources, and some randomness too

  Finally spit out the guest configuration files or dump configuration
  parameters somewhere else.