I have been reading about the halftime parameter
in scheduler configuration (qconf -ssconf) and
I understand that the value of this parameter
decides "how fast the grid engine system forgets
the information about user's consumption of
resources".
Also, I understand that that smaller value of
halftime implies that the system quickly forgets
user's consumption and as such, reduces the impact
of such (over/under) consumption.
I have the following questions:
1. From 'qconf -ssconf', I see that halflife
has a value of 168 - is this in hours, days?
If hours (or days), does this begin at a certain
time every week/month/year or is this a rolling
value (meaning, past 168 units of time)?
2. If I have a 'public usage cluster' and want
to make sure my users get a fair share of
resources (no one is more special than anyone,
since none of them shared the $$ of building
this cluster) over a period 1 year, what should
be a practical value of halflife parameter?
3. In spite of this being a 'public usage cluster',
suppose that I want to reward users who consistently
use the facility for (meaningful) computations,
compared to those who use it when they feel like it.
For e.g., Tom and Harry both use it for the exact
same duration in the month of June;
then Tom goes away to lala land for vacation in
July while Harry continues to use the system at
the same consistent rate;
When Tom and Harry both submit a bunch of jobs in
August, I'd like Harry's jobs to have higher priority
than Tom's.
What changes do I need to make to make this scenario
happen?
It'd be great if you could please point me to some resources
that I can use to learn these things. Thank you for your
time and help.
Best,
g
--
Gowtham
Information Technology Services
Michigan Technological University
(906) 487/3593
http://www.it.mtu.edu/
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
