Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Performance Improvements by adding SMP to OS5.0.6

0 views
Skip to first unread message

Don Yakubowski

unread,
May 15, 2002, 5:58:54 PM5/15/02
to
Some general questions for those with experience using SCO SMP module on
dual processor OS 5.0.6 systems.....
1) Is it at all possible to predict with any accuracy how much
performance will change
by adding SMP and a 2nd processor to a system? Since there are alot of
variables
involved relating to how the apps are written and the drivers used for
the "specific"
peripherals,it seems to me that unless something (ie. a device driver or
application)
was written to take advantage of such underlying hardware,the end user
wouldn't
notice any difference at all?
The SCO docs suggest that unless an app or driver is multithreaded,it
would run on the
base CPU,just like it would without SMP,so what would be the point in
adding SMP and
another CPU to the hardware?

2) Aside from the obvious purchase of the SMP license(why do they think
it is worth as much as the base OS anyway..?),are there other issues to be
wary of?I have noticed postings about issues with specific hdwe that works
on an NON-SMP system that suddenly has issues with SMP(which I presume is
because the drivers need to be "updated" to work on an SMP system?). We have
never felt our apps suffered from running on a single cpu system,so until
recently,haven't bothered looking at SMP.
Our apps are Cobol based so yet another variable exists,besides the OS and
hdwe under it.

--
Don Yakubowski
Tri-Comp Systems Ltd.
1-888-561-6555 ext 225
do...@tricomp.ca


Tony Lawrence

unread,
May 15, 2002, 7:01:21 PM5/15/02
to
Don Yakubowski wrote:
> Some general questions for those with experience using SCO SMP module on
> dual processor OS 5.0.6 systems.....

> The SCO docs suggest that unless an app or driver is multithreaded,it


> would run on the base CPU,just like it would without SMP,so what would be the
> point in adding SMP and another CPU to the hardware?

Multiple users running multiple processes.

>
> 2) Aside from the obvious purchase of the SMP license(why do they think
> it is worth as much as the base OS anyway..?),are there other issues to be
> wary of?I have noticed postings about issues with specific hdwe that works
> on an NON-SMP system that suddenly has issues with SMP(which I presume is
> because the drivers need to be "updated" to work on an SMP system?). We have
> never felt our apps suffered from running on a single cpu system,so until
> recently,haven't bothered looking at SMP.

What evidence do you have that your systems are cpu bound? Have you
looked at sar -q ?

--

Tony Lawrence
SCO/Linux Support Tips, How-To's, Tests and more: http://pcunix.com
Free Unix/Linux Consultants list: http://pcunix.com/consultants.html

Bela Lubkin

unread,
May 15, 2002, 8:23:14 PM5/15/02
to sco...@xenitec.on.ca
Don Yakubowski wrote:

> Some general questions for those with experience using SCO SMP module on
> dual processor OS 5.0.6 systems.....
> 1) Is it at all possible to predict with any accuracy how much
> performance will change
> by adding SMP and a 2nd processor to a system? Since there are alot of
> variables
> involved relating to how the apps are written and the drivers used for
> the "specific"
> peripherals,it seems to me that unless something (ie. a device driver or
> application)
> was written to take advantage of such underlying hardware,the end user
> wouldn't
> notice any difference at all?
> The SCO docs suggest that unless an app or driver is multithreaded,it
> would run on the
> base CPU,just like it would without SMP,so what would be the point in
> adding SMP and
> another CPU to the hardware?

It is possible to predict with any accuracy only in a few specific
cases.

If your application isn't already CPU bound, adding another CPU probably
won't help.

If your application uses one monolithic process that accumulates all the
CPU time, adding another CPU probably won't help.

If your application uses lots of cooperating processes which all consume
moderate amounts of CPU, or if different users have different sets of
CPU-consuming processes running at once, adding a CPU might help a lot.

What you read in the docs probably wasn't very well written (I'm not
sure what you found). What you need to understand is that OSR5 doesn't
have kernel-level threads support, so any one process can use only one
CPU at once (even if it uses some sort of threading library). If your
machine was dedicated to a single process calculating Pi to a billion
digits, a 2nd CPU would do nothing. But if you were running one task
calculating Pi to a billion digits and another calculating `e' to a
billion digits, both tasks might run twice as fast in a 2-CPU system,
since each would end up on a separate CPU.

>Bela<

0 new messages