The correct link is http://tinyurl.com/2w62pj

Also, the problem only apply to the opening of new windows/programs.
Programs which are already open when the problem surface continue
running: firefox, for instance, is still able to load huge pages...

Thank yo anyone able to help!

Piersante

piesse wrote:
>
> I had theseus already open, just in case, and the output of the linear
> memory map page is posted at http://tinyurl.com/2w62pj(It is too long to
> be posted here)

piesse wrote:
> The correct link is http://tinyurl.com/2w62pj
>
> Also, the problem only apply to the opening of new windows/programs.
> Programs which are already open when the problem surface continue
> running: firefox, for instance, is still able to load huge pages...
>
> Thank yo anyone able to help!
>
> Piersante
>
> piesse wrote:
>>
>> I had theseus already open, just in case, and the output of the linear
>> memory map page is posted at http://tinyurl.com/2w62pj(It is too long
>> to be posted here)

You'd be better off posting ram usage by process. The problem is that
OS/2 only has 512 MBs available by process and DLLs load into one 512 MB
space. The system uses some of this so you only have 300-400 MBs per
process. One thing that might help if you are running 4.5 or newer is to
use one of Peters builds of Firefox/Seamonkey available here,
http://pmw-warpzilla.sf.net/.
Dave

Dave Yeo wrote:
>
> You'd be better off posting ram usage by process. The problem is that

here it is:
http://tinyurl.com/2f8a8b


> OS/2 only has 512 MBs available by process and DLLs load into one 512 MB
> space. The system uses some of this so you only have 300-400 MBs per
> process. One thing that might help if you are running 4.5 or newer is to
> use one of Peters builds of Firefox/Seamonkey available here,
> http://pmw-warpzilla.sf.net/.

I already use them

Thanks,
Piersante

piesse wrote:
> Dave Yeo wrote:
>>
>> You'd be better off posting ram usage by process. The problem is that
>
> here it is:
> http://tinyurl.com/2f8a8b
>
>
>> OS/2 only has 512 MBs available by process and DLLs load into one 512
>> MB space. The system uses some of this so you only have 300-400 MBs
>> per process. One thing that might help if you are running 4.5 or newer
>> is to use one of Peters builds of Firefox/Seamonkey available here,
>> http://pmw-warpzilla.sf.net/.
>
> I already use them
>
> Thanks,
> Piersante
>
What is your virtualaddresslimit set to? It should be >= 1536
(defaults to 1024 if not present in config.sys which should be fine).
If it is greater than 1536 then I would suggest lowering it. The
only two things I can really tell myself from your Theseus report is
that you are using high memory for your Mozilla programs, and that
(while I don't believe it to be your problem) Norman is using a lot of
memory (not quite 200M total in all its processes).
Andy

Andy Willis wrote:
> What is your virtualaddresslimit set to? It should be >= 1536 (defaults

it is not set, si it should be the default


> to 1024 if not present in config.sys which should be fine). If it is
> greater than 1536 then I would suggest lowering it. The only two things
> I can really tell myself from your Theseus report is that you are using
> high memory for your Mozilla programs, and that (while I don't believe
> it to be your problem) Norman is using a lot of memory (not quite 200M
> total in all its processes).

I see in my linear map this line:

"There is 0.000M between the private and shared arenas."

Is that an indication that some program (Norman?) has a memory leak?
thanks
Piersante

Andy Willis wrote:
> piesse wrote:
>> Dave Yeo wrote:
>>>
>>> You'd be better off posting ram usage by process. The problem is that
>>
>> here it is:
>> http://tinyurl.com/2f8a8b
>>
>>
>>> OS/2 only has 512 MBs available by process and DLLs load into one 512
>>> MB space. The system uses some of this so you only have 300-400 MBs
>>> per process. One thing that might help if you are running 4.5 or
>>> newer is to use one of Peters builds of Firefox/Seamonkey available
>>> here, http://pmw-warpzilla.sf.net/.
>>
>> I already use them
>>
>> Thanks,
>> Piersante
>>
> What is your virtualaddresslimit set to? It should be >= 1536 (defaults
> to 1024 if not present in config.sys which should be fine). If it is
> greater than 1536 then I would suggest lowering it. The only two things
> I can really tell myself from your Theseus report is that you are using
> high memory for your Mozilla programs, and that (while I don't believe
> it to be your problem) Norman is using a lot of memory (not quite 200M
> total in all its processes).
> Andy
>
Correction... meant less than or equal to, not greather than or equal to.
Andy

In <5jbim5F3tm026U1@mid.individual.net>, on 08/25/2007
at 10:47 PM, piesse said:

Hi,

>I had theseus already open, just in case, and the output of the linear
>memory map page is posted at http://tinyurl.com/2w62pj(It is too long to
>be posted here)

You have not run out of address space. There's free space available, but
the lower shared areaa boundary has moved unusually low and the lower
shared arena is somewhat fragmented.

What the Linear Memory map says is that your system is pretty busy and
that Mozilla has chewed up quite a bit of private address space.

While you may not have run out of address space, you might have run out of
some limited resource that is responsible for keep track of memory
objects. This is something I don't know how to determine directly from
Theseus. I guess one could post-processing the Linear memory map and get
an estimate.

If shutting down the running apps and restarting them resolves your
problem, there's probably nothing wrong with the data structures that are
managing memory. If not, then you need to look at what's still using
memory after you shutdown everything you can. One of these might have a
leak of some sort.

One needs to keep in mind that much of the kernel internals are still
16-bit and the many of the internal ids used to keep track of things are
limited to 65K unique items. This was a lot back in the early 90's when a
typical system had a lot less physical memory. The inevitable swapping
limited the number of apps one would attempt to run concurrently. It easy
to imagine running out of available memory object ids on today's systems
with a couple GB of physical RAM.

Steven

--
--------------------------------------------------------------------------------------------
Steven Levine MR2/ICE 3.00 beta 08pre #10183
eCS/Warp/DIY/14.103a_W4 www.scoug.com irc.ca.webbnet.info #scoug (Wed 7pm PST)
--------------------------------------------------------------------------------------------

On Sat, 25 Aug 2007 20:47:43 UTC, piesse wrote:

> From time to time, my system seems to eat up all the memory.
> In this moment, for instance, if I try to open even a simple command
> window, it fails saying that there is non enough memory available :-(
>
>
>
> Shmemmon says: Available: 715 256Kb blocks = 178.8 Mb
>
> I had theseus already open, just in case, and the output of the linear
> memory map page is posted at http://tinyurl.com/2w62pj(It is too long to
> be posted here)
>
> I am unable to interpret it. Is anyone able to tell me what is going
> wrong? Is there any other check that I could do?
>
> thanks
>
> Piersante

Just a WAG, but try adding:
dllbasing=off
near the top of your CONFIG.SYS.

Hope this helps...
--
From the eComStation 1.2 of Doug Bissett
dougb007 at telus dot net
(Please make the obvious changes, to e-mail me)

Doug Bissett wrote:

>
> Just a WAG, but try adding:
> dllbasing=off
> near the top of your CONFIG.SYS.

I have it already

It just seems that the good old efficient memory management of os/2
cannot cope anymore with the memory requests of current applications
(or, seen the other way around, that current applications are not
dealing efficiently with it).

Running out of resources while still having >200M of free RAM is
disappointing.

Which is a pity.

To me, this is the more critical aspect of os/2-eCS nowadays (not having
a current java in sight being the second).

Of course, we are used to live with some difficulties. But it starts to
be disturbing.

thanks
Piersante

On Mon, 27 Aug 2007 23:20:07 UTC, piesse wrote:

> Doug Bissett wrote:
>
> >
> > Just a WAG, but try adding:
> > dllbasing=off
> > near the top of your CONFIG.SYS.
>
> I have it already
>
> It just seems that the good old efficient memory management of os/2
> cannot cope anymore with the memory requests of current applications
> (or, seen the other way around, that current applications are not
> dealing efficiently with it).
>
> Running out of resources while still having >200M of free RAM is
> disappointing.
>
> Which is a pity.
>
> To me, this is the more critical aspect of os/2-eCS nowadays (not having
> a current java in sight being the second).
>
> Of course, we are used to live with some difficulties. But it starts to
> be disturbing.
>
> thanks
> Piersante

You are, probably, running out of "shared memory space". With later
versions of OS/2, (at least the MCP based versions, with later
kernels), there is a second shared memory space, in high memory. That
is only usable by programs that are programmed to use it, and some of
them (such as the FAT32 driver) need to be told to use that, rather
than the low memory space. If you GOOGLE, you should find a number of
posts on the subject, with many suggestions as to what to do to work
around the problem.

Hope this helps...
--
From the eComStation 1.2 of Doug Bissett
dougb007 at telus dot net
(Please make the obvious changes, to e-mail me)

piesse wrote:
> From time to time, my system seems to eat up all the memory.
> In this moment, for instance, if I try to open even a simple command
> window, it fails saying that there is non enough memory available :-(

As you will know from my recent VOICE articles this applies to virtual
address space. Not physical memory.

> Shmemmon says: Available: 715 256Kb blocks = 178.8 Mb
Shmemmom measures the free virtual memory blocks of at least 256Kb in
the lower shared arena.

Usage: shmemmon interval [logfile [maxsize]]
Parameters: interval Time between samples in minutes
logfile Path and filename for logfile
maxsize Maximum size of logfile in Kb

When it measured the free 1MB blocks needed to load in virtual memory
the memory needs of large (1 MB) DLLs, shmemmom would be more pessimistic.

Theseus mentions that the free pool of memory (the pool of virtual
memory space once meant to load new memory objects in) has gone.

095A0000 x (<- moved)
095B0000


There is 0.000M between the private and shared arenas.
Shared arena starts at 095A0000, which is 149.625M.
Free memory from 095A0000 for 102.313M, which is equivalent to 1637
64K spaces.
0FBF0000
..w.....w....w.ww...w.ww.wwwwwwwwwwwwwwwwwww..www.w.w.wwwwwwwwwwwwwwwwwwwwowww
713 PMMERGE allocated it
object is 2.813M

If a new memory object is to be loaded in the shared arena (open a new
program), the loader has to find some hole elsewhere.

Scanning the theseus ouput for "Free memory from" shows that you do have
large unfragmented (virtual memory) areas (45M, 17M) satisfying the
memory needs of even much larger applications than CMD:

Free memory from 10AE0000 for 42.500M, which is equivalent to 680 64K
spaces.
13560000
..w.....w....w.ww...w.ww.wwwwwwwwwwwwwwwwwww..www.w.w.wwwwwwwwwwwwwwwwwwwwwwww
123 PMMERGE allocated it
object is 512K
Free memory from 135E0000 for 17.000M, which is equivalent to 272 64K
spaces.
146E0000
.................................................. ........o...................
1 XSERVER #0004 (shared data)
Free memory from 146F0000 for 64K.

This kind of fragmentation in the virtual memory address space cannot
explain why you cannot load a new cmd.

There must be some other resource running "out of memory" unless the
OS/2 loader is very picky about virtual address spaces to load cmd
related files (which is not my experience).

In the OS/2 2.0 memory model it could not be low private virtual memory,
as this can be given for free (only the largest low private memory
consumer can do harm. But a new cmd.exe should be tiny unlike java.exe).


> I had theseus already open, just in case, and the output of the linear
> memory map page is posted at http://tinyurl.com/2w62pj(It is too long to
> be posted here)

This was a good idea.

> I am unable to interpret it. Is anyone able to tell me what is going
> wrong? Is there any other check that I could do?
>
> thanks
>
> Piersante

After reverse engineering (looking queer) the list of running
applications, I have to conclude that you ask a lot of OS/2. I wonder if
you could achieve this- with the same hardware - also under Windows
(little change) or Linux (likely, but slow).

You have running many servers ftpd, httpd, squid, weasel, major, ncsa_au
and wksta. You forgot ffsd, syslogd, mysql and smb (;-)
And did not disable RUN=F:\OS2\SMSTART.EXE (do you need it?)

Now you are running on a dedicated server without OS/2 firewall (i
suppose because of lack of ffsd and syslogd) Mozilla and Thunderbird.
How long has your server been up?

On my Linux (SuSE)LAN server I use a crown script to reload the smbd,
nmbd and xinetd (used by imap) hourly to prevent smb and imap users
complaining.

I use in etc/crown/hourly

#! /bin/bash
/etc/init.d/xinetd restart
/etc/init.d/smb restart
/etc/init.d/nmb restart
exit


So I agree with the answer of Steven Levine, stating that - though the
shared arean is somewhat fragmented - another source must have been out
of memory preventing the cmd.exe to load. Maybe a 16 bit kernel source
may be out of memory.

Anyway, shutting down and reloading MOzilla and some other servers (like
i needed under Linux) at regular intervals might help you.


Sjoerd Visser

piesse schrieb:
> Andy Willis wrote:
>> What is your virtualaddresslimit set to? It should be >= 1536 (defaults
>
> it is not set, si it should be the default
>
>
>> to 1024 if not present in config.sys which should be fine). If it is
>> greater than 1536 then I would suggest lowering it. The only two
>> things I can really tell myself from your Theseus report is that you
>> are using high memory for your Mozilla programs, and that (while I
>> don't believe it to be your problem) Norman is using a lot of memory
>> (not quite 200M total in all its processes).
>
> I see in my linear map this line:
>
> "There is 0.000M between the private and shared arenas."
>
> Is that an indication that some program (Norman?) has a memory leak?
> thanks
> Piersante

True, Mozilla makes the private memory arena hit the shared memory arena
in the low memory region (< 512 MB). On the other hand, it still has
room and uses private memory arena in the high memory region (>512 MB).
Your linear address map also reports 72 MB of free shared memory in the
low memory region which is enough (and plenty of free shared memory
space in the high memory region but this only helps if DLLs are enabled
to load into the high memory region).
None of your processes seem to take an extraordinary amount of memory
(example: your Mozilla takes 149 MB of private memory address space in
the low memory region, here, Seamonkey takes 576 MB of private memory
address space !).

As I can tell from the end of your address range (0xC0000000 = 3072 MB)
you are using VIRTUALADDRESSLIMIT=3072.
The problem is that this decreases the system memory arena (memory
address range that can only be used by the system and that starts at
VIRTUALADDRESSLIMIT). System memory is used for things like video
aperture, PCI address space, device driver code and data segments, OS
internal data structures etc.
If the system cannot allocate system memory things like you are
describing can happen: impossible to start new processes etc.
Try decreasing VIRTUALADDRESSLIMIT to let's say 2048 and see if it helps.
Also set DLLBASING=OFF (without the SET statement !) in config.sys as
you seem to have big memory holes between the loaded DLLs which could
indicate that they load on their hardcoded base addresses (which however
is unnecessary and only gives a minimal performance benefit when they
are initially loaded into memory).

Finally, you can mark code and data segments of DLLs to load in high
memory address space rather than in low memory address space (use the
tool exehdr.exe and run it with the /HIGHMEM parameter against the
DLLs). However that might lead to problems, so have a backup handy. The
DLLs that are good candidates are those that are listed in your shared
address space in low memory region (<512 MB), in particular, all DLLs
that come with Mozilla.

Note: using address space is not equal to using that much memory. You
can allocate address space without also allocating (committing memory).
Lacking memory is prevented/recovered by swapping, lacking address space
is fatal.



Lars

Lars Erdmann wrote:
> As I can tell from the end of your address range (0xC0000000 = 3072 MB)
> you are using VIRTUALADDRESSLIMIT=3072.


Actually no, or at least VIRTUALLADDRESSLIMIT is currently unset (it
doesn't occur in Config.sys or in the output of a SET command)

> Also set DLLBASING=OFF (without the SET statement !) in config.sys as
> you seem to have big memory holes between the loaded DLLs which could

Thanks, I will try this
>
> Finally, you can mark code and data segments of DLLs to load in high
> memory address space rather than in low memory address space (use the
> tool exehdr.exe and run it with the /HIGHMEM parameter against the
> DLLs). However that might lead to problems, so have a backup handy. The
> DLLs that are good candidates are those that are listed in your shared
> address space in low memory region (<512 MB), in particular, all DLLs
> that come with Mozilla.

Maybe I will try it later, although in effect the problem has occurred
once more and was only cured by shutting down both Mozilla and
Thunderbird (mozilla alone wasn't sufficient.
I must probably have to care not to live them open for too long.


>
> Note: using address space is not equal to using that much memory. You
> can allocate address space without also allocating (committing memory).

thanks for all your help!

Piersante

Sjoerd Visser wrote:
> After reverse engineering (looking queer) the list of running
> applications, I have to conclude that you ask a lot of OS/2. I wonder if
> you could achieve this- with the same hardware - also under Windows
> (little change) or Linux (likely, but slow).


This is exactly why I use os/2 ;-)

>
> You have running many servers ftpd, httpd, squid, weasel, major, ncsa_au
> and wksta. You forgot ffsd, syslogd, mysql and smb (;-)

Actually, mysql is usually there, but it was down for some reason.

> And did not disable RUN=F:\OS2\SMSTART.EXE (do you need it?)

I don't even know what it is, except that it takes care of Ffst.exe
Erlogger.exe Dmisl.exe remoterr.exe

>
> Now you are running on a dedicated server without OS/2 firewall (i
> suppose because of lack of ffsd and syslogd) Mozilla and Thunderbird.
> How long has your server been up?

In the good old times it used to stay up until ibm released a new
fixpack and I had to reboot :-)
Now it freezes from time to time.


>
> Anyway, shutting down and reloading MOzilla and some other servers (like
> i needed under Linux) at regular intervals might help you.
>

I have found that I must close both Mozilla and Thunderbird to recover
space between the private and shared arenas

Thanks a lot!
Piersante

On Mon, 10 Sep 2007 10:47:11 +0200, piesse wrote:

>> After reverse engineering (looking queer) the list of running
>> applications, I have to conclude that you ask a lot of OS/2. I wonder
>> if you could achieve this- with the same hardware - also under Windows
>> (little change) or Linux (likely, but slow).
>
>
> This is exactly why I use os/2 ;-)

If running the two big 'DEs' I'd agree, but on older hardware a lighter
Linux desktop environment might be faster. XFce is pretty good and
improving all the time and I know some people who prefer it over Gnome
and KDE, even on newer hardware.

Wayne
--
Registered Linux user #375994
http://www.geocities.jp/rondonko/

piesse wrote:
> Sjoerd Visser wrote:
>> After reverse engineering (looking queer) the list of running
>> applications, I have to conclude that you ask a lot of OS/2. I wonder
>> if you could achieve this- with the same hardware - also under Windows
>> (little change) or Linux (likely, but slow).
>
>
> This is exactly why I use os/2 ;-)
>

Exactly.
--
Posted with OS/2 Warp 4.52
and Sea Monkey 1.5a

On Mon, 10 Sep 2007 08:25:41 UTC in comp.os.os2.misc, piesse
wrote:

> Actually no, or at least VIRTUALLADDRESSLIMIT is currently unset (it
> doesn't occur in Config.sys or in the output of a SET command)

It wouldn't be listed in the output of SET as it is not a SET command. You just
specify VIRTUALADDRESSLIMIT= not SET VIRTUALADDRESSLIMIT=


--
Trevor Hemsley, Brighton, UK
Trevor dot Hemsley at ntlworld dot com

Trevor Hemsley wrote:
>
> It wouldn't be listed in the output of SET as it is not a SET command. You just
> specify VIRTUALADDRESSLIMIT= not SET VIRTUALADDRESSLIMIT=
>
ahah, you are right, it was there! There was a REM note saying that I
had set it because of apache (that was probably with the early v.2 releases)

I have no idea why grep failed to spot it (probably a mistyping)
I changed it and will see

thanks!

Piersante

Here in comp.os.os2.misc, Wayne spake unto us, saying:

>On Mon, 10 Sep 2007 10:47:11 +0200, piesse wrote:
>
>>> After reverse engineering (looking queer) the list of running
>>> applications, I have to conclude that you ask a lot of OS/2. I wonder
>>> if you could achieve this- with the same hardware - also under Windows
>>> (little change) or Linux (likely, but slow).
>>
>> This is exactly why I use os/2 ;-)
>
>If running the two big 'DEs' I'd agree, but on older hardware a lighter
>Linux desktop environment might be faster. XFce is pretty good and
>improving all the time and I know some people who prefer it over Gnome
>and KDE, even on newer hardware.

Even smaller LiveCD distros like Puppy and DSL are slower than OS/2 on
old enough hardware. With newer hardware both are probably lightning
fast, but you can see a difference on a PPro. :-)

--
-Rich Steiner >>>---> http://www.visi.com/~rsteiner >>>---> Mableton, GA USA
Mainframe/Unix bit twiddler by day, OS/2+Linux+DOS hobbyist by night.
WARNING: I've seen FIELDATA FORTRAN V and I know how to use it!
The Theorem Theorem: If If, Then Then.