[Beowulf] scheduler policy design

[Tim Cutts]

On 25 Apr 2007, at 5:43 pm, Bogdan Costescu wrote:

> On Wed, 25 Apr 2007, Tim Cutts wrote:
>
>> With our queueing system that forces the job into a lower priority  
>> queue.
>
> Just to be sure that we are talking about the same thing: this is a  
> value which is set by the user to the maximum allowed, not one  
> automatically used by the scheduler because none was provided.


> Then how do you distinguish between these careless users and those  
> super-smart who try to squeeze every drop of the CPU time that they  
> can get and actually have jobs that run very close to the maximum  
> time allowed ? (the question doesn't apply only to time, but could  
> be rephrased for any other resource)
> I don't know how probable such a mixture of users is, but I did  
> encounter it ;-)

I think this all reflects my original assertion right back at the  
beginning of the thread; optimising for a variety of different  
requirements simultaneously is basically impossible.   If you have a  
highly variable job mix, about all you can do is try to keep the CPUs  
as busy as possible, and employ some limits to keep non-CPU resources  
as much under control as you can, and live with the fact that your  
scheduling will rarely be optimal.

Tim