WLM and PRM

[Ninad_1]

Hi guys,

I want to clear few doubts I have on the practical use of WLM and PRM from those who are using it.
Suppose I have 2 vPars in a rp7420 with 2 cells in a single nPar using PA8900 as follows
vPar#1 6 CPU, 32 GB RAM (For production environment)
vPar#2 4 CPU, 16 GB RAM (For development environment)
Now using WLM and PRM if I configure workload groups for the 2 vPars and say set a rule that if usage of production server goes above 90% then it should get CPU from development vPar, then how does this actually happen.
I understand from the docs that every 30 secs the wlmd checks the status from all workload groups and takes necessary actions.
Now suppose the CPU load on prodcution server is continuously fluctuating between say 89 and 91 %, then will there be continuous adjustments of CPU between the vPars ? Doesnt this have any adverse effect or any overhead ?
What happens in practical scenario ?
I can think of an analogy which I remember of a concept called chattering of a relay (electromagnetic relay swicth) which switches on and off depending on the potential difference between its poles, now when the potential difference is near the limit when it should switch on/of there used to be chatter due to frequent turn on and off of the relay contacts. Does something similar happen in WLM/PRM ? How to take care of this, does it have any implications in practical terms.
Please guide with your experiences. Is it worth thinking of using WLM and PRM.

Thanks,
Ninad

[Steven E. Protter · Accepted solution]

Shalom Ninad,

I'm not an advanced user.

Now suppose the CPU load on prodcution server is continuously fluctuating between say 89 and 91 %, then will there be continuous adjustments of CPU between the vPars ? Doesnt this have any adverse effect or any overhead ?
What happens in practical scenario ?

CPU's are supposed to flucate, they are called upon to do different amounts of work at different times.

No adverse effect.

Practical scenario: In a situation where there is a need to direct resources on a moderate to busy system to get important work done before less important work, PRM is extremely useful. I recall using to make sure a VERY important report to a VP got done faster as it waded in a sea of reports on a moderately busy server.

Is it useful on a nearly idle system? Nope. But it can help you prepare your system to be very efficient and do important work first when some nice user hits the gas pedal.

SEP

[Ninad_1]

Steve,
So even though there will be a continuous deallocation and allocation of CPU resources between vPars, you say that this does not have any adverse effect - right.
Any other caveats ? Any useful tips/things I need to keep in mind ?

Thanks,
Ninad