CPU Reduction Cost

I concur with Callandor, you need to understand how (if) resources are charged for.

Some places there are no real charges for jobs, so when you reduce the resource required for a process all you are doing is prolonging the time until you need to upgrade, if you need to upgrade.

Would the cost be the same if done on a Unix system?  Probably not.  Most mainframe shops that do real charging charge for CPU, Memory, and I/O.  However for Unix system if they do charge it was for CPU, they typically don't charge (or measure) for memory or I/O resources.

Hello Callandor, qiltjr,
Basically i need to arrive at a generic idea on how it is done.

In MF itself it shows

Initial CPU Time: 1.856, Elapsed: 1.897, Phys I/O: 19983, QUnits: 51538, Monetory Cost: 0.4124

What are these? How are these used? How is it done similarly for Unix?

If i can arrive at a formulae with X dollars usage per CPU minute usage or with respect to peak reduction or whatever then i can correlate to Daily, Weekly and Monthly job and come with an yearly savings.

That X value what is being paid i can check with my systems and accouts person.

Regards.
Hari.

"If i can arrive at a formulae with X dollars usage per CPU minute usage or with respect to peak reduction or whatever then i can correlate to Daily, Weekly and Monthly job and come with an yearly savings."

Even so, what if your cpu's are over or under-spec? What if you're running single threaded apps on a system that is designed for multiple threads or vice-versa?

If a user request takes 10 seconds to service, is this deemed as acceptable? What would it take to get to 5 seconds? More hardware? Software architecture redesign? Both?

I agree will Callandor and giltjr in saying that perhaps your barking up the wrong tree. This sounds like a maths homework assignment rather than something that would happen in the real world.

Hi,

what you basically need is -

- Purchase cost of your hardware divided by its expected lifetime

- Maintenance cost per time period

- Operation cost per time period (energy, cooling, ...)

- Infrastructure cost per time period (floor space, network, other cabling, ...)

- Human resources cost per time period (operators, systems managers, programmers, ...)

- Other cost per time period not mentioned above (whatever you need to spend to keep your IT running)

As Callandor wrote, your accounting dept. should be able to tell you those values.

Now you should have your total cost per time period.

Divide this cost by the amount of those resources consumed in that time period you'd like to chargeback (CPU, I/O or the like).

This will give you a certain amount of money per unit of resources consumed (e.g. CPU minute).

HTH

wmp

--> Initial CPU Time: 1.856, Elapsed: 1.897, Phys I/O: 19983, QUnits: 51538, Monetory Cost: 0.4124

Initial CPU time : number of CPU seconds it took to complete the work.  The number of CPU seconds will vary each time you run the job.  The number of CPU seconds is not the same for each model of IBM mainframe, so CPU seconds on model is not the same as CPU seconds on model Y.

Elsapsed: The "wall clock time", that is the number of seconds it took to run the job based on your wrist watch.  This time also can vary from job run to job run.

Physical I/O: The number of reads or write to any and all files this job used.

QUnits: I have no idea, this is something that your shop may have made up based on something else.

Cost:  You need to ask somebody how your company bases its charges.

The list that woolmilkporc  provide are the items that need to be accounted for in order to come up with the cost.  No matter what type of system.

Typically when saying "mainframe" people do me IBM mainframes running z/OS, z/VSE, or z/VM.  As I stated before, in these environments different computer resources (CPU, memory, and I/O) are allocated different costs.  In Unix (and Windows) it typically based on walk clock run time and/or CPU.  However, today you can run Linux or Solaris on an IBM mainframe and in those cases the charge back system would be the same as running on a non-mainframe system.

Its not the hardware that gives you the break down of resources consumed, its the OS.  Distributed OS's can provide the same information as traditional mainframe OSs, but most shops just use CPU and elapsed.

QU is DB records retrieved (Query units) - i.e either ROW or BLOB.
I have no idea what is charged in your case, you can reduce QUNITS by retrieving more objects at once.

Q Initial CPU Time: 1.856, Elapsed: 1.897, Phys I/O: 19983, QUnits: 51538, Monetory Cost: 0.4124
A Where did these values come from? What command generated the values?

All these values i had got it from Mainframe system after running the job. Nothing related to DB it is job run specific.

Regards.
Hari.

One thing is that the resources used by a job normally varies from run to run based on various factors.  Such as the number of "records" processed and how busy the system is at the time the job runs.