Author Topic: Autowrangling behavior  (Read 1637 times)

mim

  • Full Member
  • ***
  • Posts: 14
Autowrangling behavior
« on: July 16, 2013, 01:54:12 PM »
Is there a possibility to tweak the autowrangling behavior?

For instance we've got 24core as well as 12core nodes in our farm. can i tell qube to "prefer" the faster machines when assigning jobs to idle machines?


greetings
mim

jburk

  • Administrator
  • *****
  • Posts: 484
Re: Autowrangling behavior
« Reply #1 on: July 16, 2013, 06:56:27 PM »
This isn't so much tweaking the auto-wrangling behavior (which handles what to do when jobs aren't running properly) so much as you'd like to change the default dispatch behavior.

You can control which hosts the supervisor will prefer to send jobs to with a supervisor parameter supervisor_default_hostorder.  This is a new feature in Qube 6.5, and is used in the supervisor's qb.conf.

To instruct the supervisor to send jobs to workers with the most processors installed, you would set this value to:

host.processors.total

to send jobs to a worker with the most processors installed and the fewest in use:

host.processors.total,-host.processors.used

This parameter is documented here:
http://docs.pipelinefx.com/display/QUB065/supervisor_default_hostorder

If you're not running Qube 6.5, you can set the host_order on a per-job basis, and can set it as a default in the various submission UI's in WranglerView.  You will need to tick the 'expert mode' box at the bottom of the UI to expose this control.

mim

  • Full Member
  • ***
  • Posts: 14
Re: Autowrangling behavior
« Reply #2 on: July 17, 2013, 07:49:12 AM »
thanks alot for your quick answer!
since we're running 6.4-4 at the moment i'll adapt the studio prefs