On Fri, 18 Jul 1997, David Randall wrote:

>                     ------QUANTUM-------
> QUEUE  BASE  LIMIT  MIN    MAX    ACTUAL  BOOST  TIMESLICE
> -----  ----  -----  ---    ---    ------  -----  ---------
>  CQ    152    200   1      2000   18      DECAY    200
>  DQ    202    238   2000   2000   2000    DECAY    200
>  EQ    160    238   1      200    25      DECAY    200
>
> I have tried oscillating the DQ and EQ but it seemed to make
> things worse (fast performance for a period then nothing
> for several seconds)

 A small overlap in the C and D queues is a great way to go.  If you want
the batch processes to grab a little more CPU as they continue to run
then set the D queue to oscillate.  You will see it execute in what could
be called a herky-jerky fashion but it will grab more cpu over the course
of the job.  Newly introduced, and long running, batch jobs will grab
more CPU than long running C queue processes.  Simple C queue tasks will
always get top priority (after system processes).

 You can also overlap your E queue.  If you push the top of the E queue
into the C queue region and leave the E queue on decay, then anyone
working in the E queue who performs a simple task will see good response
time.  Anyone doing a larger task will wait........

 Sort of like the following:

 CQ --start---






                DQ --start---
                                 EQ --start---
 CQ --decay---





                DQ --oscill--



                                 EQ --decay---

 Slide the scale according to your specific needs and your observations
as to how it helps/hinders performance.  Your mileage may vary.

 If you've got some $$$ in the budget... might be worth looking into HP's
workload manager.

Good luck,
Jim

--------------------------------------------------------------------------
James R. Alton
Systems Analyst
Mohawk College                        phone: (905) 575-2281
135 Fennell Ave. W.                     Fax: (905) 575-2302
Hamilton, Ont., Canada L8N 3T2        Email: [log in to unmask]
--------------------------------------------------------------------------