buildworld: Strange memory problem

buildworld: Strange memory problem

Post by Steven G. Kar » Wed, 25 Jun 2003 01:49:47





Quote:> OK, I got 128MB of new memory for my old box (Cyrix MII). Memtest-86
> runs ok for 2 hours, no problem at all. The whole system (FBSD 4.8
> Release) runs rockstable too with the new memory.

> When I try to run buildworld, I occasionally get linker errors:

> /usr/libexec/elf/ld: final link failed: File too large

> I can continue the make (-D NOCLEAN) and I get the same error 2-3
> times again, but finally it builds fine.

Did you increase the amount of swap space when you added the
128 MB?  This seems to be an out-of-swap space problem.
What does swapinfo report?

--
Steve
http://troutmask.apl.washington.edu/~kargl/

 
 
 

buildworld: Strange memory problem

Post by David Bolto » Fri, 27 Jun 2003 03:30:45




>> Did you increase the amount of swap space when you added the
>> 128 MB?  This seems to be an out-of-swap space problem.
>> What does swapinfo report?

> No, there is altogether 192 MB memory and 130 MB swap space. During the

Swap can be in more than one place.  On my old Pentium I have a 64M swap
partition on each of my two HDs, and the kernel recognizes both; 'top' reports
'Swap: 128M Total'.  If you have a partition that you can shave a piece off and
reformat, that would give you the needed additional swap, and confirm (or not)
whether the limited swap space was indeed your problem.

Best regards,
David
--
Little minds are tamed and subdued by misfortune,
but great minds rise above it.
- Washington Irving

 
 
 

buildworld: Strange memory problem

Post by Steven G. Kar » Fri, 27 Jun 2003 05:14:20





>> reformat, that would give you the needed additional swap, and confirm (or not)
>> whether the limited swap space was indeed your problem.

> Again: there is no swap space problem, as I can buildworld with the
> same swap space and less RAM.

I don't think you can make this statement unilaterally.  Some
structures internal to the kernel may be dynamically allocated
depending on the amount of physical ram and swap space.  

You can test the insufficient swap space theory by using
vnconfig on 4.x or mdconfig on 5.x to create additional
file-backed swap space.

Quote:> The problem is a sophisticated RAM problem, as it happens only
> occasionally.

Are the memory modules matched (i.e., both PC100 or PC133 memory)?

Quote:> Actually I can live with that, as a buildworld is needed only every
> few months. The whole system seems to run fine otherwise. Still, there
> may be a slight chance that strange memory problem may finally lead to
> some corruption.

Hopefully, you don't do any serious computations on this machine
such as design aircraft, medical treatment protocal planning,
control nuclear reactors. ...  :-O

--
Steve
http://troutmask.apl.washington.edu/~kargl/

 
 
 

buildworld: Strange memory problem

Post by David Bolto » Fri, 27 Jun 2003 07:17:18




>> reformat, that would give you the needed additional swap, and confirm (or not)
>> whether the limited swap space was indeed your problem.

> Again: there is no swap space problem, as I can buildworld with the
> same swap space and less RAM.

Buildworld failed when you had a 192:130 RAM to swap ratio; it worked when you
cut it to 64:130.  Maybe someone can clarify this: are there processes that
need a minimum of 1:1?  Wasn't this ratio at one time a required minimum?

--
I find the great thing in this world is not so much where we stand, as in what
direction we are moving.
- Oliver Wendell Holmes

 
 
 

buildworld: Strange memory problem

Post by Michael Sierchi » Fri, 27 Jun 2003 08:02:40



> Buildworld failed when you had a 192:130 RAM to swap ratio; it worked when you
> cut it to 64:130.  Maybe someone can clarify this: are there processes that
> need a minimum of 1:1?  Wasn't this ratio at one time a required minimum?

You don't need swap at all -- NO_SWAPPING is a kernel option,
and of course there are MIBs to control the behavior of swap,
including vm.swap_enabled=0

Ancient paging algorithms preferred at least twice as much
disk swap partition space as physical RAM.  It matters little
now, though running out of virtual memory still has the same
consequences (processes get killed).

 
 
 

buildworld: Strange memory problem

Post by Kris Kennawa » Fri, 27 Jun 2003 11:01:14




>> Buildworld failed when you had a 192:130 RAM to swap ratio; it worked when you
>> cut it to 64:130.  Maybe someone can clarify this: are there processes that
>> need a minimum of 1:1?  Wasn't this ratio at one time a required minimum?

> You don't need swap at all -- NO_SWAPPING is a kernel option,
> and of course there are MIBs to control the behavior of swap,
> including vm.swap_enabled=0

NO_SWAPPING doesn't do what you probably think it does.  Read the
description in NOTES.

Quote:> Ancient paging algorithms preferred at least twice as much
> disk swap partition space as physical RAM.

This seems like FUD..FreeBSD's VM is hardly ancient.

Kris

 
 
 

buildworld: Strange memory problem

Post by Michael Sierchi » Fri, 27 Jun 2003 11:23:01



> NO_SWAPPING doesn't do what you probably think it does.  Read the
> description in NOTES.

NO_SWAPPING does precisely what I think it does, but thanks for
playing the Amazing Kreskin.  At the county fair you'd be giving
away lotsa teddy bears.  It's a define for vm_pageout.c et al.
Most of the swap code is omitted from the kernel if it's set.

Quote:>>Ancient paging algorithms preferred at least twice as much
>>disk swap partition space as physical RAM.

> This seems like FUD..FreeBSD's VM is hardly ancient.

You need remedial reading lessons.  I said that it's no longer
an issue, but that ancient (okay, <= BSD 4.3) paging algorithms relied
on swap being twice RAM.  That's just a historical fact.  But then,
I can still name the members of that band Paul McCartney had before
Wings, so I've been around a while.

What I'm saying is counter-FUD -- in fact, you can still read
erroneous advice such as:

        Your swap partitions should total at least twice the size of
        the amount of RAM memory you have installed in your machine.

        - http://www.vmunix.com/fbsd-book/book.phtml

There was indeed a good reason for this once.  It would be impossible
to use FreeBSD as an embedded OS if this were really true.  Fortunately,
a read-only root and MFS for /var works just fine.  No swap.  Ass too
high, run too fast.

 
 
 

1. Is "make -k buildworld" equal to "make buildworld" the first time run?

Hi, I am galactic_war and I am back. Just a quick question:

 Last time I did "make buildworld" and "make installworld", I generated
/usr/obj
which I burned to CD along with /usr/src. After I install FreeBSD on a
different
computer, I copied from this CD the /usr/src and /usr/obj to this new
computer. I did a
"make -k buildworld" with the "-k" option, hoping that some object file
would not need to
be compiled again. But it didn't seem to be any faster.

 I just wonder, since I didn't have any error doing "make buildworld" on
the first
computer, would "make -k buildworld" save me some practical time by
preserving the object files in the /usr/obj directory? Are "make -k" and
"make" the same if they are run the first time?

 Thank you for any suggestions?

--
vector sigma

2. Printing to multiple printers at the same time

3. Is "make -k buildworld" equal to "make buildworld" the first time run

4. named problem?

5. Is "make -k buildworld" equal to "make buildworld" the first time

6. Installing without root-authority?

7. strange compiler signal 11 when doing make buildworld

8. Suggestion for SCSI controller

9. Strange problem with 64MB memory

10. Strange memory problem

11. Strange memory/speed problem.

12. Strange memory problem on Toshiba Satellite Laptop

13. A strange memory problem.