Q: System w/ limited 256 kB cache, 16MB->32MB = slower linux.

Q: System w/ limited 256 kB cache, 16MB->32MB = slower linux.

Post by Worawut Wisutmethangoo » Tue, 11 Jun 1996 04:00:00



Sorry, I know this question has been asked many times.

My system is 486-66MHz with "256kB cache" running Linux kernel 1.2.6.
I have 16 MB RAM and am trying to add 16 MB more. The results is Linux
is noticably slower, whereas Windows 3.1 runs faster than before.

I have also read, in this newsgroup some time ago, that the problem is
due to not enough cache, which should be at least 16kB per 1MB of RAM.
Am I right about this?

The problem I have is that my motherboard can have maximum cache of
256 kB, and there are no sockets for more cache.  Right now, the only
options I know are to :

 1) Return the memory and be content w/ 16 MB
 2) Buy new motherboard for the memory and cache

Are there any other ways that you can suggest? E.g., Will upgrading
linux kernel help? Or is there any cache card that I can add onto
the motherboard?

I would really appreciate any pointer or suggestion.

Regards,
Worawut Wisutmethangoon


 
 
 

Q: System w/ limited 256 kB cache, 16MB->32MB = slower linux.

Post by Worawut Wisutmethangoo » Tue, 11 Jun 1996 04:00:00


I like to ask some more questions. In the manual of my motherboard,
it says that 256kB of cache can handle main memory upto 32 MB.
This probably is true for Windows 3.1 as I see no speed penalty.
What makes it different in Linux? Could anyone please explain?

Regards,
Worawut Wisutmethangoon


 
 
 

Q: System w/ limited 256 kB cache, 16MB->32MB = slower linux.

Post by David Gaudi » Wed, 12 Jun 1996 04:00:00




Quote:>Sorry, I know this question has been asked many times.

>My system is 486-66MHz with "256kB cache" running Linux kernel 1.2.6.    
>I have 16 MB RAM and am trying to add 16 MB more. The results is Linux
>is noticably slower, whereas Windows 3.1 runs faster than before.

>I have also read, in this newsgroup some time ago, that the problem is
>due to not enough cache, which should be at least 16kB per 1MB of RAM.
>Am I right about this?

Could be, but what I read is that on some motherboards there's a jumper
(or setup option) to set if you want to cache all memory, not just the lower
16 meg.
 
 
 

Q: System w/ limited 256 kB cache, 16MB->32MB = slower linux.

Post by David Gaudi » Wed, 12 Jun 1996 04:00:00




Quote:>I like to ask some more questions. In the manual of my motherboard,
>it says that 256kB of cache can handle main memory upto 32 MB.
>This probably is true for Windows 3.1 as I see no speed penalty.
>What makes it different in Linux? Could anyone please explain?

To see a speed penalty, you'd have to actually have something running in
the upper half of memory.  Perhaps Linux puts something up there and
win3.1 doesn't?
 
 
 

1. GUS + BusLogic BT-445S + STB Pegasus + DX2/66 + 256 kB cache = trouble

I previously wrote

The situation is even weirder. I can solve the problem by any *one*
of the following:

        - disable external cache
        - use 128 kB cache instead of 256 kB
        - replace the DX2/66 with a DX/33
        - remove the GUS (gravis ultrasound)
        - replace the VL-bus BT-445S with an ISA-bus AHA-1542B
        - replace the VL-bus STB Pegasus with an ISA-bus S3-801 vga card

Things that would not help:

        - replacing cache SRAM
        - replacing DRAM SIMMS
        - replacing 20 ns TAG SRAM with a 15 ns chip
        - changing any AMI BIOS setup option other than disabling ext. cache
        - adding VL-bus waitstates
        - slot positions (though 1 VL slot does not support busmastering)

I am now waiting for a newer revision of my Buslogic card, but
I'm beginning to think the problem is in the motherboard;
things are just getting too fast, I guess.

There is also another problem that happens *only* when
using 128 kB cache: the infamous signal 11 internal gcc error
when building a Linux kernel.

Before this all happened, I had another cache problem:
Linux would crash randomly, mainly in the swapper task,
unless I had disabled the external cache.
This was solved by building a kernel with -m386 instead of -m486.
This worked, but I could never find out the theory behind this.

This is already my second motherboard; the first one would
appear totally dead with the STB Pegasus on it.

For the record, here's my hardware setup:

        motherboard: DX2/66, SiS 461 VL chipset, 3 VL slots, AMI BIOS
        16 MB 70 ns DRAM
        256 kB 20 ns SRAM, 32 kB 15 ns TAG RAM
        BusLogic BT-445S VL SCSI host adapter rev D, BIOS 4.72, f/w 3.36
        STB Pegasus VL S3-928 vga card (S3-928 is D-stepping)
        Gravis Ultrasound (don't have the revision number)
        no-name ISA-bus 2x16550 1xparallel I/O card
--

S-mail : Laan Copes van Cattenburch 70, 2585 GD The Hague, The Netherlands
Phone  : +31 70 3542302
Fax    : +31 70 3512837

2. Berkeley db v1.85 compile problem.

3. 32MB Slower than 16MB

4. Netscape Communicator Hangs

5. 32MB is slower than 16MB?

6. How to get mouse pop-up:s disapper?

7. ISDN at 256 kb ... howto ?

8. MY MISTAKE.

9. Workshop 4.2 -> 5.0 -> 6.0 slower and slower...

10. Creative 3D Blaster Geforce 256 Annihilator Pro 32MB AGP

11. 256 threads limit on linux

12. HELP! Linux 256 file descriptors Limit!

13. Linux 99pl8A dies with 32Mb ram, 16Mb works fine