SEO

I am building a server that has 4 gig of RAM using OpenServer 5.0.5. The
large amount of RAM is for I/O buffers for a DBMS.

I built a 6 gig swap area, thinking a smaller swap area might be limiting
how much ram the OS can see.

I boot with mem=1m-16m,16m-4092m/n/p to force OS to see all the RAM. The
message shows all the RAM is found.

Still the boot up messages tell me I have a
"mem: total = 787388k" with
"kernel = 18384k, user = 769004k"

Have I run into some kind of undocumented limit?

Also, the man for mem says that, by default, RAM above 16m is not DMA. Is
this a restriction or a default? Could it be affecting my problem?

Thank you in advance.

--
Frank

Nature, and cats, abhor a vacuum.

Frank Mabrey wrote:

> I am building a server that has 4 gig of RAM using OpenServer 5.0.5. The
> large amount of RAM is for I/O buffers for a DBMS.
>
> I built a 6 gig swap area, thinking a smaller swap area might be limiting
> how much ram the OS can see.
>
> I boot with mem=1m-16m,16m-4092m/n/p to force OS to see all the RAM. The
> message shows all the RAM is found.
>
> Still the boot up messages tell me I have a
> "mem: total = 787388k" with
> "kernel = 18384k, user = 769004k"
>
> Have I run into some kind of undocumented limit?

Are you sure you're running 5.0.5? The amount it's reporting is how
much you would see with OpenServer 5.0.0 or 5.0.2, which had a 768MB
memory limit. Also, 5.0.0 and 5.0.2 would only see 512MB unless you
forced the issue with a "mem=" bootstring like you show; 5.0.4 and later
will see the extra memory by default.

The amount of RAM OpenServer sees is not at all limited by swap. The
amount you can _use_ may be limited slightly if you have very little
swap (it varies greatly depending on how your applications uses memory,
but under typical conditions you will need your swap to be at least 1/4
as big as your RAM to be able to use all of RAM).

No single swap area can be larger than 4GB, though you can have multiple
4GB swap areas. Before 5.0.6 behavior with >4GB swap was a bit
unpredictable, so you should definitely reduce that to 4GB.

> Also, the man for mem says that, by default, RAM above 16m is not DMA. Is
> this a restriction or a default? Could it be affecting my problem?

Memory above 16MB cannot be reached by the old ISA DMA controller (which
is still used by e.g. standard floppy drives). This is a hardware
limitation; it's also of very little importance in 2003, when most
devices that use DMA are PCI devices that have access to the full 4GB
address space.

>Bela<

In article <xNUqb.140443$Fm2.116918@attbi_s04> "Frank Mabrey" <SKIPTHISfrank@mabrey.com> writes:
$Have I run into some kind of undocumented limit?

Hmm ... the documentation for my copy of 5.0.5 says 512 MB.
But the FAQ (http://aplawrence.com/SCOFAQ/scotec3.html#max_ram)
says starting in 5.0.4 OSR5 can use up to 4 GB. I'm guessing that
someone forgot to update the docs when they increased this limit.

What happens if you omit the mem bootstring? Usually, you
only use such a bootstring if boot isn't correctly enumerating your
RAM.

$Also, the man for mem says that, by default, RAM above 16m is not DMA. Is
$this a restriction or a default? Could it be affecting my problem?

This is an archaic limit for ISA cards. The ISA bus was designed
for the IBM PC AT, which used an 80286 CPU. The 286 can only access
16 MB of RAM, and therefore the ISA bus only has enough address lines
(24) to access 16 MB of RAM. (Some microchannel cards have a similar
limit.) The PCI bus has full 32-bit addressing capability and
so it can access a full 4 GB of RAM.

This doesn't affect how much memory can be recognized; it only
affects how I/O is performed. The operating system will automatically
work around this, if needed, by allocating some buffers in the first
16 MB and copying data to/from them. This reduces performance when
accessing such a device. Many PCs from the last several years don't
even have ISA slots any more, though I think the floppy controller
probably still suffers from this limit (but as it's a low-performance
device that's no big deal).
--
Stephen M. Dunn <stephen@stevedunn.ca>
>>>----------------> http://www.stevedunn.ca/ <----------------<<<
------------------------------------------------------------------
Say hi to my cat -- http://www.stevedunn.ca/photos/toby/

In article <20031107224257.GG14056@sco.com>,
Bela Lubkin <belal@sco.com> wrote:
>Frank Mabrey wrote:

>> I am building a server that has 4 gig of RAM using OpenServer
>> 5.0.5. The large amount of RAM is for I/O buffers for a DBMS.

>> I built a 6 gig swap area, thinking a smaller swap area might
>> be limiting how much ram the OS can see.

>> I boot with mem=1m-16m,16m-4092m/n/p to force OS to see all
>> the RAM. The message shows all the RAM is found.

>> Still the boot up messages tell me I have a
>> "mem: total = 787388k" with
>> "kernel = 18384k, user = 769004k"

>> Have I run into some kind of undocumented limit?

>Are you sure you're running 5.0.5? The amount it's reporting is
>how much you would see with OpenServer 5.0.0 or 5.0.2, which had
>a 768MB memory limit. Also, 5.0.0 and 5.0.2 would only see 512MB
>unless you forced the issue with a "mem=" bootstring like you
>show; 5.0.4 and later will see the extra memory by default.

>The amount of RAM OpenServer sees is not at all limited by swap. The
>amount you can _use_ may be limited slightly if you have very little
>swap (it varies greatly depending on how your applications uses memory,
>but under typical conditions you will need your swap to be at least 1/4
>as big as your RAM to be able to use all of RAM).

On the standard Intel chips [not the Itaniums] isn't memory limited
to 4GB unless the OS support PAE - Physical Address Extension?
And if that's correct does the latest OSR5 support it. Just
curious.

Bill


--
Bill Vermillion - bv @ wjv . com

In article <Ho0M57.8zp@wjv.com>, Bill Vermillion <bv@wjv.com> wrote:
>In article <20031107224257.GG14056@sco.com>,
>On the standard Intel chips [not the Itaniums] isn't memory limited
>to 4GB unless the OS support PAE - Physical Address Extension?

Yes.

>And if that's correct does the latest OSR5 support it.

No, OpenServer doesn't use PAE. UnixWare does.

John
--
John DuBois spcecdt@armory.com KC6QKZ/AE http://www.armory.com/~spcecdt/

(Frank Mabrey) 07.11.03 in /comp/unix/sco/misc:

>I am building a server that has 4 gig of RAM using OpenServer 5.0.5.
>The large amount of RAM is for I/O buffers for a DBMS.

>I built a 6 gig swap area, thinking a smaller swap area might be
>limiting how much ram the OS can see.

>I boot with mem=1m-16m,16m-4092m/n/p to force OS to see all the RAM.
>The message shows all the RAM is found.

>Still the boot up messages tell me I have a
>"mem: total = 787388k" with
>"kernel = 18384k, user = 769004k"

>Have I run into some kind of undocumented limit?

Yepp.

Long time ago there were patches to fix this rediculous/surprising limit
(for a (once) hi-end OS).
But SCO decided to withdraw them, stating that they (the patches)
cause problems...

To get arround it you have to upgrade to at least 5.0.6 AFAIK.
But you stay fixed at a 2GB file size limit...

Maybe you like to have a look at "A.B.I." (see Tony's webpages)
and can host you application under linux?

Maybe it would be less "expensive" to migrade to Linux and Postgresql?
And: What do think will happen to SCO next year?
I remember very good that they once stopped development on
OSR5 some time ago and resumed it with 5.0.7 mainly to introduce
a new paying model...

Rainer Zocholl wrote:

> (Frank Mabrey) 07.11.03 in /comp/unix/sco/misc:
>
> >I am building a server that has 4 gig of RAM using OpenServer 5.0.5.
> >The large amount of RAM is for I/O buffers for a DBMS.
>
> >I built a 6 gig swap area, thinking a smaller swap area might be
> >limiting how much ram the OS can see.
>
> >I boot with mem=1m-16m,16m-4092m/n/p to force OS to see all the RAM.
> >The message shows all the RAM is found.
>
> >Still the boot up messages tell me I have a
> >"mem: total = 787388k" with
> >"kernel = 18384k, user = 769004k"
>
> >Have I run into some kind of undocumented limit?
>
> Yepp.
>
> Long time ago there were patches to fix this rediculous/surprising limit
> (for a (once) hi-end OS).
> But SCO decided to withdraw them, stating that they (the patches)
> cause problems...

The patches were for OpenServer 5.0.0 and 5.0.2 only, and they _did_
cause problems, they weren't anywhere near as stable as 5.0.4. Starting
with OSR504 (shipped in mid-1997) OpenServer supports a full 4GB. (The
patches for 500/502 only supported 2GB). Why are you upset that you can
no longer get poorly functional patches for OSes that were superseded 6
1/2 years ago?

> To get arround it you have to upgrade to at least 5.0.6 AFAIK.

5.0.4. The original poster hasn't responded to my questioning his
version number, which should not be 5.0.5 if he's topping out at 768MB.

> But you stay fixed at a 2GB file size limit...

True.

> Maybe you like to have a look at "A.B.I." (see Tony's webpages)
> and can host you application under linux?
>
> Maybe it would be less "expensive" to migrade to Linux and Postgresql?
> And: What do think will happen to SCO next year?
> I remember very good that they once stopped development on
> OSR5 some time ago and resumed it with 5.0.7 mainly to introduce
> a new paying model...

OpenServer development has not stopped at any time in the last 10 years.
The company has had different ambitions for the OS and different numbers
of people working on it, but never stopped. Right now, development is
ramping up as we prepare to deliver the features promised at SCO Forum
2003.

>Bela<

Zounds! The disk I was using was 5.0.2. (red face)

The messages from the boot never said what version was being booted. Of
course the media disk doesn't either. Thanks to your suggestion I finally
used "uname -X" to find my mistake.

Thank you for your timely and comprehensive reply. I'd send you a cookie
but internet cookies aren't really all that tasty.

Frank

"Bela Lubkin" <belal@sco.com> wrote in message
news:20031107224257.GG14056@sco.com...
> Frank Mabrey wrote:
>
> > I am building a server that has 4 gig of RAM using OpenServer 5.0.5.
The
> > large amount of RAM is for I/O buffers for a DBMS.
> >
> > I built a 6 gig swap area, thinking a smaller swap area might be
limiting
> > how much ram the OS can see.
> >
> > I boot with mem=1m-16m,16m-4092m/n/p to force OS to see all the RAM.
The
> > message shows all the RAM is found.
> >
> > Still the boot up messages tell me I have a
> > "mem: total = 787388k" with
> > "kernel = 18384k, user = 769004k"
> >
> > Have I run into some kind of undocumented limit?
>
> Are you sure you're running 5.0.5? The amount it's reporting is how
> much you would see with OpenServer 5.0.0 or 5.0.2, which had a 768MB
> memory limit. Also, 5.0.0 and 5.0.2 would only see 512MB unless you
> forced the issue with a "mem=" bootstring like you show; 5.0.4 and later
> will see the extra memory by default.
>
> The amount of RAM OpenServer sees is not at all limited by swap. The
> amount you can _use_ may be limited slightly if you have very little
> swap (it varies greatly depending on how your applications uses memory,
> but under typical conditions you will need your swap to be at least 1/4
> as big as your RAM to be able to use all of RAM).
>
> No single swap area can be larger than 4GB, though you can have multiple
> 4GB swap areas. Before 5.0.6 behavior with >4GB swap was a bit
> unpredictable, so you should definitely reduce that to 4GB.
>
> > Also, the man for mem says that, by default, RAM above 16m is not DMA.
Is
> > this a restriction or a default? Could it be affecting my problem?
>
> Memory above 16MB cannot be reached by the old ISA DMA controller (which
> is still used by e.g. standard floppy drives). This is a hardware
> limitation; it's also of very little importance in 2003, when most
> devices that use DMA are PCI devices that have access to the full 4GB
> address space.
>
> >Bela<

(Bela Lubkin) 08.11.03 in /comp/unix/sco/misc:

>Rainer Zocholl wrote:

>> Maybe you like to have a look at "A.B.I." (see Tony's webpages)
>> and can host you application under linux?
>>
>> Maybe it would be less "expensive" to migrade to Linux and
>> Postgresql? And: What do think will happen to SCO next year?
>> I remember very good that they once stopped development on
>> OSR5 some time ago and resumed it with 5.0.7 mainly to introduce
>> a new paying model...

>OpenServer development has not stopped at any time in the
>last 10 years.

Yepp "stopped" was a too hard word.

But reduced to a minimum. (In 2000 you must have had the feeling
that they have decided (internal) to let OSR die.)

>The company has had different ambitions for the OS and
>different numbers of people working on it, but never stopped.

Ok, call it "different ambition" ;-)
I have the impression that they did only the development
that can't be avoided, meaning fixing "Class A" bugs.

>Right now, development is ramping up as we prepare to deliver
>the features promised at SCO Forum 2003.

Approx. 2 years ago i only found on the hardware lists (HCL) machines which
were at least 2 years not available anymore...no modern machines.

Too i would assume that a "dedicated" high end server operating system
should support 1Gbit ethernet in 2000.
3 years ago the only way to use 1Gb was an (illegal) driver from
the 3com site.
Intel 1Gb was not supported at all.
The 1Gb drivers by SCO are released no in Aug 2003! wow... approx. 5 Years
development for a simple ethernet driver. But well,
you are right, it is not stopped...;-)

The OS had a lot of trouble with modern fast CPUs.
But that's clear: The HCL tests stopped in (approx) 1998 at 200MHz,
it's obvious that they never have tested > 400GHz machines
until the customers began to complain...

What date has the latest skunkware CD? AFAIK it's still 2000?


I would not call that "development" i would call it "stagnation" ;-)

In article <8xWq6zR6gjB@zocki.toppoint.de>,
Rainer Zocholl <UseNet-Posting-Nospam-74308-@zocki.toppoint.de> wrote:
>(Bela Lubkin) 08.11.03 in /comp/unix/sco/misc:
>
>>Rainer Zocholl wrote:

>>> Maybe you like to have a look at "A.B.I." (see Tony's webpages)
>>> and can host you application under linux?

>>> Maybe it would be less "expensive" to migrade to Linux and
>>> Postgresql? And: What do think will happen to SCO next year?
>>> I remember very good that they once stopped development on
>>> OSR5 some time ago and resumed it with 5.0.7 mainly to introduce
>>> a new paying model...

>>OpenServer development has not stopped at any time in the
>>last 10 years.

>Yepp "stopped" was a too hard word.

>But reduced to a minimum. (In 2000 you must have had the feeling
>that they have decided (internal) to let OSR die.)

I have no idea what they plan, but SCO was the last company to
release a SysVR4 Unix - and that was Unixware. The OSR5 is
based on SysVR3 but with nay SysVR4 things moved back to it, but
there are some things that make no sense [from my POV] to even
attempt moving them from a VR4 to a VR3.

>>Right now, development is ramping up as we prepare to deliver
>>the features promised at SCO Forum 2003.

>Approx. 2 years ago i only found on the hardware lists (HCL)
>machines which were at least 2 years not available anymore...no
>modern machines.

>Too i would assume that a "dedicated" high end server operating
>system should support 1Gbit ethernet in 2000. 3 years ago the
>only way to use 1Gb was an (illegal) driver from the 3com site.
>Intel 1Gb was not supported at all. The 1Gb drivers by SCO are
>released no in Aug 2003! wow... approx. 5 Years development for
>a simple ethernet driver. But well, you are right, it is not
>stopped...;-)

>The OS had a lot of trouble with modern fast CPUs.
>But that's clear: The HCL tests stopped in (approx) 1998 at 200MHz,
>it's obvious that they never have tested > 400GHz machines
>until the customers began to complain...

When some of the limits of OSR5 became too much for a client I had
on that platform I migrated them into the Unixware 7. Limits went
away, performanc went up. They got big enough to hire anther
person internally so I don't go back there anymore.

Moving from a 300MHz world into a Dual Xeon with faster drives
took one major weekly process from 2 hours to under 10 minutes.
The performance was not all from the HW upgrades, but a more
efficient OS.

Bill

--
Bill Vermillion - bv @ wjv . com