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 Historical content remains viewable.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

Unixware 7.1 memory

0 views
Skip to first unread message

David Jetter

unread,
Sep 24, 2001, 8:19:39 AM9/24/01
to
I have been asked about a Unixware 7.1 system that the operator complains
that it runs slow quite often. She wants to upgrade the memory since memory
is cheap right now. With UW 7.1, what should we check to confirm the memory
usage or system performance?

After memory is installed, should the Unixware be tweaked for the additional
memory?

Thanks,
David Jetter


Pat Welch

unread,
Sep 24, 2001, 5:01:24 PM9/24/01
to

Adding more memory is usually a no-brainer to any Unix system. Depending
on your particular version of UW7 there may be a limit of 4GB ram
imposed.

But, UW7 unfortunately does very little 'auto' tuning on it's own,
unlike OS5.

The easiest way for non-sar gurus (sar is a performance analysis tool)
to tune for better performance is to get a 3rd party tool like sarcheck
(www.sarcheck.com), which distills the cryptic sar output into plain
english with recommendations.

There are other tools that in effect perform auto tuning for you, but I
found them to cause more problems than they solved - at least on OS5.

--
----------------------------------------------------
Pat Welch, UBB Computer Services, a WCS Affiliate
Caldera Authorized Partner
Unix/Linux/Windows/Hardware Sales/Support
(209) 745-1401 Fax: (413) 714-2833
Nationwide pager: (800) 608-7122
E-mail: pat...@inreach.com
Hunt down and KILL anyone involved in NY/DC attacks!
----------------------------------------------------

Robert Lipe

unread,
Sep 24, 2001, 6:46:47 PM9/24/01
to
On Mon, 24 Sep 2001 21:01:24 GMT, Pat Welch <pat...@inreach.com> wrote:

>But, UW7 unfortunately does very little 'auto' tuning on it's own,
>unlike OS5.

Well, that's true in a left-handed way.

Systems of the SVR4 lineage don't *need* as much kernel tuning. They're
highly dynamic and can efficiently shift resources - such as memory - around
on demand instead of requiring the system administrator or some program
to figure out that "ooooh, you need more hash queues; let me turn up that
number, build you a new kernel, and remind you to reboot". You don't have
to decide at kernel configure/build (or boot) times that "I want 18.73
Megabytes dedicated to disk buffers even though I know that memory can
never be used by user applications". The system doles out pages where
they're needed and doesn't require the admin to turn knobs.


More to answer the original question for UW7/OU8, check out 'rtpm'. It
shows you where resources (cpus and memory) are being spent in a realtime-ish
format.


Brian K. White

unread,
Sep 24, 2001, 7:42:35 PM9/24/01
to
> Adding more memory is usually a no-brainer to any Unix system. Depending
> on your particular version of UW7 there may be a limit of 4GB ram
> imposed.
>
> But, UW7 unfortunately does very little 'auto' tuning on it's own,
> unlike OS5.

I thought it was the other way around.

--
Brian K. White -- br...@aljex.com -- http://www.aljex.com/bkw/
+++++[>+++[>+++++>+++++++<<-]<-]>>+.>.+++++.+++++++.-.[>+<---]>++.
filePro BBx Linux SCO Prosper/FACTS AutoCAD #callahans Satriani

Tom Parsons

unread,
Sep 25, 2001, 6:11:11 AM9/25/01
to sco...@xenitec.on.ca
Brian K. White enscribed:

| > Adding more memory is usually a no-brainer to any Unix system. Depending
| > on your particular version of UW7 there may be a limit of 4GB ram
| > imposed.
| >
| > But, UW7 unfortunately does very little 'auto' tuning on it's own,
| > unlike OS5.
|
| I thought it was the other way around.

Wrong.

OpenServer has always done a fair amount of dynamic kernel turning but
with not with the sophistication of UnixWare.
for most

--
==========================================================================
Tom Parsons t...@tegan.com
==========================================================================

David Jetter

unread,
Sep 25, 2001, 9:11:49 AM9/25/01
to

"Tom Parsons" <c...@tegan.com> wrote in message
news:2001092505...@tegan.com...

> Brian K. White enscribed:
> | > Adding more memory is usually a no-brainer to any Unix system.
Depending
> | > on your particular version of UW7 there may be a limit of 4GB ram
> | > imposed.
> | >
> | > But, UW7 unfortunately does very little 'auto' tuning on it's own,
> | > unlike OS5.
> |
> | I thought it was the other way around.
>
> Wrong.
>
> OpenServer has always done a fair amount of dynamic kernel turning but
> with not with the sophistication of UnixWare.
> for most

Well what is the verdict, does UW auto-tune? They are getting sarcheck for,
hopefully, a little more insight into the system.

Thanks to all!

David Jetter


Tom Parsons

unread,
Sep 25, 2001, 11:20:57 AM9/25/01
to sco...@xenitec.on.ca
David Jetter enscribed:

Did you read Robert Lipe's answer? For good reason, Robert has more than
passing knowledge of UnixWare and certainly much more than I (or anyone
else on this list).

Ben Rosenthal

unread,
Sep 25, 2001, 3:31:53 PM9/25/01
to
Robert Lipe wrote:

> On Mon, 24 Sep 2001 21:01:24 GMT, Pat Welch <pat...@inreach.com> wrote:
>
> >But, UW7 unfortunately does very little 'auto' tuning on it's own,
> >unlike OS5.
>
> Well, that's true in a left-handed way.
>

On behalf of the left-handed people who read this newsgroup, I hope it's true in
a _good_ way. Remember left-handed is right brained. 8-))

>
> Systems of the SVR4 lineage don't *need* as much kernel tuning. They're
> highly dynamic and can efficiently shift resources - such as memory - around
> on demand instead of requiring the system administrator or some program
> to figure out that "ooooh, you need more hash queues; let me turn up that
> number, build you a new kernel, and remind you to reboot". You don't have
> to decide at kernel configure/build (or boot) times that "I want 18.73
> Megabytes dedicated to disk buffers even though I know that memory can
> never be used by user applications". The system doles out pages where
> they're needed and doesn't require the admin to turn knobs.
>
> More to answer the original question for UW7/OU8, check out 'rtpm'. It
> shows you where resources (cpus and memory) are being spent in a realtime-ish
> format.

So does this mean that UW7/OU8 does a better job at dynamic kernel tuning then
OSVR5?

Ben Rosenthal


Robert Lipe

unread,
Sep 25, 2001, 8:50:58 PM9/25/01
to
On Tue, 25 Sep 2001 19:31:53 GMT, Ben Rosenthal <bcr...@attglobal.net> wrote:
>Robert Lipe wrote:
>
>> On Mon, 24 Sep 2001 21:01:24 GMT, Pat Welch <pat...@inreach.com> wrote:
>>
>> >But, UW7 unfortunately does very little 'auto' tuning on it's own,
>> >unlike OS5.
>>
>> Well, that's true in a left-handed way.
>>
>
>On behalf of the left-handed people who read this newsgroup, I hope it's true in
>a _good_ way. Remember left-handed is right brained. 8-))

It's just a saying. I'm not trying to pick a fight with anybody...
The phrase means things that are true, but in a counterinuitive way.

>> Systems of the SVR4 lineage don't *need* as much kernel tuning. They're
>> highly dynamic and can efficiently shift resources - such as memory - around
>> on demand instead of requiring the system administrator or some program
>> to figure out that "ooooh, you need more hash queues; let me turn up that
>

> So does this mean that UW7/OU8 does a better job at dynamic kernel

> tuning than OSVR5?

Reduced to a boolean quantity, I'd have to say "yes".

There are many fewer knobs for the admin to turn. The reality is that the
OS can know how better to tune things than the admin will. And it can
adjust to things on the fly - without kernel relinks; without reboots - so
the system can make good use of the resources *at this moment*. The page
pool can look very different on an idle system at 3 a.m. then when it
starts the backup than when 750 users log in in the morning.

I could also bore you with things about reduced paging, improved cache
coherency, less "dead" memory, and lots of other things, but I'll stick
with my answer of "yes, the SVR5 kernel does a better job of dynamic
tuning."

Signed,
A Senior S/W Engr in the SVR5 Kernel Development Team

Matt Schalit

unread,
Sep 26, 2001, 1:39:40 AM9/26/01
to
On Mon, 24 Sep 2001 08:19:39 -0400, "David Jetter" <dje...@springmail.com> wrote:

>I have been asked about a Unixware 7.1 system that the operator complains
>that it runs slow quite often. She wants to upgrade the memory since memory
>is cheap right now. With UW 7.1, what should we check to confirm the memory
>usage or system performance?

There's a few paths. People have mentioned them,
including rtpm, sar, and common sense. Look for any
yellow and red values in rtpm, other than %run,
which should not be red at 0%.

Often it takes quite a bit of common sense, and
I'm smiling as I type that.

Take for example the "eager" user who often runs
programs that hog the cpu, like gcc, or there could
be a runaway process.

It doesn't take a degree to install as much ram
as you can afford, as many processors, and the fastest
scsi disks. Disks, as you may know, take milliseconds
to access, as versus nanoseconds to access ram. There
are some nice processors out there, too, these days.

>After memory is installed, should the Unixware be tweaked
>for the additional memory?

The simple answer is, "No, Sir." Your more likely to hit
other limits that you might change for peace of mind and
whatnot, like NPROC, the maximum number of processes.

Check out the Uw7 FAQ question, "What are the important
tunable parameters I might adjust like MAXUP?"

http://www.zenez.com/tmp/scouw7faq/cache/183.html

Regards,
Matt

>Thanks,
>David Jetter

Hey you take care now,
Matt


David Jetter

unread,
Sep 28, 2001, 10:55:59 PM9/28/01
to
"Tom Parsons" <c...@tegan.com> wrote in message
news:2001092510...@tegan.com...<snip>
<snip>

> |
> | Well what is the verdict, does UW auto-tune? They are getting sarcheck
for,
> | hopefully, a little more insight into the system.
>
> Did you read Robert Lipe's answer? For good reason, Robert has more than
> passing knowledge of UnixWare and certainly much more than I (or anyone
> else on this list).

When Robert posted his reply, I did read it, 11 hrs.after I posed this
question.

Thanks to all
David

David Jetter

unread,
Sep 28, 2001, 10:59:21 PM9/28/01
to
"Matt Schalit" <msch...@pacbell.net> wrote in message
news:3bb15cdc...@news.sf.sbcglobal.net...

Thanks for the straight forward answer.

David


0 new messages