Re: [Condor-users] Power Management

[Dan Bradley <dan@xxxxxxxxxxxx>]

On 8/11/11 9:11 AM, Ian Cottam wrote:
> Some questions and help sought with Power Management (Wake-on-LAN, etc.).
>
> I had some early success with a simple test on a small pool with Condor
> 7.4, and following pretty much what the manual says.
>
> We are now trying on more machines in our main pool; we see:
> - Condor sleeping the machine when jobs are still active.

What is your HIBERNATE expression?

If you can get a startd log that covers the time period in which an 
event like this happens, please send it to condor-admin@xxxxxxxxxxxx  It 
would be most useful with D_FULLDEBUG turned on in STARTD_DEBUG.

> - Sometimes the LAN light stops flickering altogether (but we may have
> fixed that one).
> - If a user is logged-in but goes inactive for a while, the machine is
> shutdown. If he or she walks away, when the next user (or Condor) wakes up
> the Linux (red hat) box the user is still logged in.

What would you prefer---that users are logged out when the machine 
hibernates, or that Condor does not hibernate the machine when users are 
logged in?  The former sounds like it would require modification to the 
system hibernation process.  The latter could possibly be achieved by 
writing a startd cron script that advertises how many users are logged 
in and then referring to this in your hibernate expression.

> Some specific questions:
> 1. Would we see major improvement by going to 7.6?

I'm not aware of any big improvements related to power management in 7.6 
vs. 7.4.
> 2. Would anyone like to share their condor_config.local lines that they
> use successfully with power management?
> 3. We use dynamic slots. We have one slot per machine that is then
> dynamically partitioned. Is this likely to be impacting power management?
Not that I am aware.

--Dan