Am 15.07.2011 um 12:44 schrieb Dave Love:
> Should we trust something so alcoholic? Thanks for offering it round,
> though.
>
> Denis Demidov <[email protected]> writes:
>
>>> Look at the "Green Computing" thread from April 29, 2011. There are
>>> some good comments in the follow ups.
>>>
>>
>> Thanks for the tip. I've modified the script slightly, now it tries to power
>> up only enough nodes to meet resource requests of pending jobs. There is
>> also a way to manage only specified subgroup of nodes.
>
> Good, though PE requests may be relevant too, where there are islands of
> connectivity, for instance. In the end, you probably think that GE
> itself should be sorting that out... I've never delved into how exactly
> the Hedeby/SDM stuff interacts with GE, but it might be instructive.
Yep, I was curious too. As they say that they reserve also shut-down nodes'
slots until they have enough for a parallel job and switch them on again when
it's worth to start them.
As Fritz mentioned its delevolpment stopped at Oracle.
> In addition to the comments about power control (see also powerman for
> an abstraction over that, but it has security issues):
>
> * It needs a copyright statement with a suitable licence ("free as in
> freedom, not just free as in beer"?).
>
> * I'd be inclined to run such things as a global load sensor, which
> mainly means modifying the main loop. I think it's a bit neater to be
> started and stopped with the qmaster.
>
> * I can't remember if I said so before, but the best way to power off a
> node while avoiding races (between the qhost output and poweroff
> taking full effect) is probably to submit a job to do it which grabs
> exclusive access. Unfortunately the detail are potentially
Or disable the queue instances on this machine, still no job there => shut it
down, enable it again.
-- Reuti
> site-dependent. Otherwise, the ssh command could do a soft stop of
> the execd service and then check that there isn't a job (shepherd) on
> the node before powering it off, but that also depends on OS details.
>
> As a heads-up: I recently had another go at getting the results of
> <http://www.jisc.ac.uk/whatwedo/programmes/greeningict/technical/supercomputers.aspx>
> on doing such things with PBS (/Torque?). It's expected to be adaptable
> to GE, and seems to have considerable operational experience behind it,
> but may need significant work on the implementation. I hope Tolis will
> be able to sort it out soon, but I don't want to hassle hard-pressed
> fellow sysmans.
> _______________________________________________
> users mailing list
> [email protected]
> https://gridengine.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
