Re: buffers vs. pages vs. kernel speed

Andrew Purtell (apurtell@ieinc.com)
Fri, 13 Jun 1997 12:23:08 -0400


I applied the same patch to pre-2.0.31 and then tried to do
with mem=8M:

dd if=/dev/zero of=/dev/ram0 bs=1k count=4096

The machine locked hard. No messages, no response to keypresses.
The hardware was a Pentium 100. There was 64M of swap available,
but the system never touched the disk after I hit Enter. And
/proc/meminfo showed 6M free before I attempted this.

I admit that this is extreme abuse, but I've never had any
2.0.x lock on me this way before and I often create 4MB ramdisks
temporarily on small 8MB 386 systems. This is a test I use to
insure I can do that before I upgrade the kernel.

On a brighter note, I can untar a kernel tarball now without
exhausting all available RAM on those same 386 systems now with
this patch. That's definitely an improvement over 2.0.30 and
pre-2.0.31.

Petri Kaukasoina wrote:
> With that patch on pre-patch-2.0.31-2 this kind of command:
>
> gunzip -c largefile.gz | egrep 'regexp'
>
> made the system rather unresponsive. Without the patch the command
> does not affect the responsiveness.
>
> I have a 66 MHz 486, 16 megs of ram. The 'largefile' was on a local
> ide disk and the size of the (compressed) file was 17 megs. Nothing
> mentionable running at the same time.

-- 
Andrew Purtell                  phone: (617) 272-9700
Network Administrator           fax  : (617) 272-9300
Intelligent Environments        email: apurtell@ieinc.com
Burlington, MA 01803 USA               akp@tiac.net