SEO

Bottom line: We can't get Enterprise on a Multi-cpu system to perform
better than Workgroup on a single CPU...

Here is the Hardware configuration:
Brand new IBM AIX eServer p5 550 with 4 CPUs, 8GB ram and a RAID 10
disk system. (To test with single CPU, we dealocate 3 CPUs and remove
them from configuration. Works perfectly to simulate single CPU.)

The benchmark:
When we submit the tests programs on the database (shared-memory, not
remote clients), the system basically stays IDLE... The disk won't go
over 35% busy, while TOTAL CPU usage will stay around 25-30 % busy
(looks like only 1 cpu running at any given time?). Only when APW&BIW
do something will the system be like 60% busy... But most of the time,
it's idling at 60-70%!!!

BTW, each test program updates DIFFERENT tables on the database; so
record locking is not the problem.

We can undestand that Workgroup should use only one processor, but what
about Enterprise? We played with all parameters found in documentation
(-B -spin -semsets, you name it! etc.). Actually, -spin will make
things worst if we set a too low value. But NEVER have we seen any
benefits of Entreprise with 4cpu over Workgroup with 1cpu.

Nothing will make full use of available hardware ressources. WHY?

Any help?? PLEASE!

Thanks.

More information:

Our benchmark uses 6 processes doing extensive read/write operations,
each on a different table of a single database to avoid any record lock
waits. Also, before running a test, we reboot the system to flush out
cache (Progress and OS).

Here are some of the numbers (in seconds):

Workgroup 1 CPU: 47.88
Workgroup 4 CPU: 86.72
Workgroup 4 CPU with Progress binded to one processor: 53.96
Enterprise 1 CPU: 51.70
Enterprise 4 CPU: 57.95

Doing the same with read-only processes give similar results: Workgroup
with single CPU being the fastest!!!

HOW CAN THIS BE?? Why doesn't Enterprise use all the available CPUs?

"Prog" <prog@mike.kouest.com> wrote in
news:1127242743.763882.207440@g49g2000cwa.googlegr oups.com:

> More information:
>
> Our benchmark uses 6 processes doing extensive read/write operations,
> each on a different table of a single database to avoid any record
> lock waits. Also, before running a test, we reboot the system to flush
> out cache (Progress and OS).
>
> Here are some of the numbers (in seconds):
>
> Workgroup 1 CPU: 47.88
> Workgroup 4 CPU: 86.72
> Workgroup 4 CPU with Progress binded to one processor: 53.96
> Enterprise 1 CPU: 51.70
> Enterprise 4 CPU: 57.95
>
> Doing the same with read-only processes give similar results:
> Workgroup with single CPU being the fastest!!!
>
> HOW CAN THIS BE?? Why doesn't Enterprise use all the available CPUs?
>
>

Progress client processes are single threaded.. Load the system with
1000 users and you will see the difference..

For some good articles and testing programs, go to
www.greenfieldtech.com and look at the articles by Tom Bascom..

Kevin