How to write a program that can stress memory subsystem as much as kernel recompile?

How to write a program that can stress memory subsystem as much as kernel recompile?

Post by Hartmut Niema » Sat, 09 Aug 1997 04:00:00




Quote:>Hello,
>    Does anybody have any programs or ideas on how to create a
>program that can stress the memory subsystem as much as the kernel
>linux recompile?
>Basically, I have a machine with some dodgy memory and it always
>gets a SIG 11..  Changing memory fixes it..  Basically, how do I
>create a program that can somehow emulate what  the compile process
>is doing.  I created a program that allocates a lot of 5 byte
>nodes and creates a linked list.  I then do parity checks on the
>whole linked list (about 20 megabytes in size)..  No problem detected.
>But gcc always fails...
>Any help appreciated..  I need to test over 200 machines and I can't afford
>to recompile and install linux on each one.

DOS/WINxx boxes? Then maybe check memtest86-1.3 on your favourite sunsite
mirror. It is not perfect though.
Or buy a cheap SCSI disk in an external case and a cheap SCSI card,
install Linux on it and just plug the hd into the system under test.
Or make them boot over NFS with a boot disk or loadlin.

All boxes with linux? Then
If you want to stress a box like the linux kernel compile does, why do
you not want to actually compile the kernel?
Of course you do not have to install it after compiling!.

If you run the script from the sig11 faq and have a box running linux
kernel makes all night long, and then diff the logs, you will find
flakey memory easily.

And if you want to be faster than an overnight kernel compile:
forget it. You want to catch problems that occur maybe once a day or
once a week.
BTW: run these tests over weekend.
My memory worked as long as the sun was not shinig hot onto the case.
It failed if it did only between 1pm and 3pm ...

Have a look at
http://cip2.e-technik.uni-erlangen.de:8080/hyplan/niemann/linux.html
and the SIG11 FAQ link there.

Quote:>(We caught a subtle problem with a specific build of Kingston memory
>chips on an HP PC).
>George

Hartmut.
 
 
 

1. How to write a program that can stress memory subsystem as much as kernel recompile?


[snip]

How about trying the MemTest-86 program?  It's worked on a memory
problem I had (which was actually my fault - 70ns RAM when the BIOS
thought it was 60ns).  Just dd it onto a floppy and boot with it in
the drive.

The LSM file for the version I have:

Title:          MemTest-86 - X86 memory test
Version:        1.1
Entered-date:   28May96
Description:    Memtest-86 is very thorough, stand alone memory test for
                386,486,586 systems.
Keywords:       testing memory


Primary-site:   sunsite.unc.edu /pub/Linux/system/Misc
                501     memtest86-1.1.lsm
                15921   memtest86-1.1.tgz
Alternate-site: N/A
Original-site:  N/A
Platform:      
Copying-policy: GPL

--

"[UNIX] appears to have the inside track on being the replacement for
  CP/M on the largest microcomputers (e.g. those based on 68000...)"

2. Term program help????

3. Need help on chosing kernel memory option, when recompile a new kernel

4. NEC 2Xi and Infomagic

5. Iwill XA100 & signal 11

6. AWE Wavetable drivers

7. How much memory do I need to recompile the kernel?

8. recompiling programs for new (1.3.4*) kernels

9. need to recompile ftape: recompile kernel too?

10. how to write program to access some memory address?

11. How to read/write kernel memory in my driver of Linux 2.2/2.4?