PRM Heirachical groups

[ted diment]

I am using heirachical FSS prm groups and I am finding that the child groups aren't inheriting their parent's cpu allocation.
e.g
group1 10 shares
group2 10 shares (parent group)
group2/child1 5 shares
group2/child2 5 shares

If I run a cpu hungry job in both group1 and group2/child1 I get a 66/33 split when I expected 50/50. if this a fault ?

any help appreciated !
Ted
HP PRM C.02.02 (20031217_140443) hpux_11.00 prmmonitor

[melvyn burnard]

Just to clear the split up, you have two groups, 1 and 2, which each get 10 shares.
Equally split this gives you 50% per group.
In group 2 you then have two sub-groups, each having 5 shares. This means they will each get 50% of the parent group, or 25% each (using the basic figures you have quoted.)

So if anything I would expect to see a 50/25 split.

One thing to bear in mind though is that the allocations will be allowed to go over their allowance if there is capacity available.
One way to test this is to turn CPU capping on and see what occurs then

[ted diment]

Melvyn,
thanks for the reply. Your description of the groups is correct.

Are you saying that the 2 child groups can only get upto 50% of their parent's cpu allocation each, even if only only 1 of them is being used. i.e. 1 child group can't inherit all it's parent's cpu allocation. It can only access it's own proportion of that.

I'll try it with capping on.

ps. I am using a perl script which will max out the cpu's on it's own to make sure that the prm groups I run it in are "filled".

thanks
ted

[melvyn burnard]

Ted, that is correct, each sub-group allocation is a relative portion of it's parent group's total allocation, hence if you had 5 sub-grous, each would get 20% of the parent's allocation if they each had the same allocation.

[ted diment]

Melvyn
the distiction I'm making is that if only 1 of those 5 sub groups was active would you expect it to get 100% of the parent's cpu allocation or 20% ?
(I hope that makes sense)

ted

[melvyn burnard]

unless cpu capping is on, then that subgroup could get even more than the parent group allocation, unless the system is under stress.


Also ensure you have patched PRM and the system, the PRM patch is PHSS_30985

[ted diment]

yes, assuming that another cpu hungry script is running in another PRM group (but not a sibling of this one), would you expect the child to get 1/5th (20%) of it's parent's allocation or 100% of it's paremt's allocation ?