[Re-sending as plain text. My bad. Been a long time since I
participated in a mailing list...]
Hi Peter:
I implemented the following approach for my customers and it works
well. The specifics can vary a bit depending on your particular
situation, but here's what I do:
1. Create a new DOUBLE complex value called "slowness".
2. Assign a value of slowness to each host as a host resource
attribute. The fastest set of hosts get a value of 0. The next fastest
set of hosts get a value of 1, and so on.
3. Use slowness in the load formula so that hosts with greater
"slowness" have a higher load and are thus filled after faster hosts.
The exact load formula depends on other components of the load formula.
For example, if generic, non-speed-sensitive load formula is "cpu", then
a good speed-sensitive load formula is "cpu+100*slowness". In this
case, "cpu" varies from 0 to 100, so the factor of 100 ensures that we
fill all slots of the fastest available speed class before we move on to
the next slower speed class.
Best,
Bob
On 10/14/2015 3:21 PM, Peter van Heusden wrote:
Hi there
I have a heterogeneous cluster with several generations (spanning
about 5 years) of CPU deployed. I'd like to have SGE (6.2u5) choose
fill up faster CPUs before using the slower CPUs.
How can I configure the cluster to do this?
Thanks!
Peter
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users
