[ On Monday, May 7, 2001 at 11:16:26 (-0400), Steven Grunza wrote: ]
> Subject: mount_mfs questions
>
>    I'm running NetBSD-1.5 on a P4 with 1GB of memory.  I keep getting "file 
> system full" messages when /tmp overflows rebuilding the locate 
> database.  Since I typically only use 100 to 200 MB of physical memory 
> while running, I'm considering creating a 256MB MFS drive mounted as 
> /tmp.

Personally I'd go for at least 500MB MFS /tmp on a machine like that --
and make sure that I have at least 1GB of swap too (hopefully striped
across 4 or more spindles).

Though capturing a crash dump on such a large machine might be more time
consuming than it's worth.  Crash dumps and savecore runs on even my 192MB
machine are slower than I'm usually willing to tolerate....

On this 192MB machine I have a 256MB MFS /tmp and I can fill it at least
half full (eg. when unpacking a tar file) without causing too much
thrashing.  In fact I've found that using it for unpacking tar files and
then doing things like "cvs import" across the result is more efficient
than even using separate spindles for the job, even when there's lots of
other stuff running on the machine.  Of course I have four swap
partitions of 500MB each too, each on UltraFAST/WIDE drives.  I learned
a long long time ago, back when I was driving 1-4 MIPS machines with 4MB
or less of RAM way beyond their capabilities that there's nothing better
for improving things than having lots of fast spindles to swap to (other
than of course having far more RAM than you can ever use! ;-).

One you move to 1.6 (or -current) the unified buffer cache may change
your memory utilisation and depending on your job mix may require you to
readjust some sizes.  I'm still quite happy with my over-sized MFS
because I don't use /tmp very often, at least not for anything of any
size, but when I do it is really for something that's of a very
temporary nature that needs to run quite fast such as the the "cvs
import" jobs I mentioned.  I do find when watching with systat that UBC
alone causes far more paging activity overall than the MFS did, but with
fast swap devices it's not really painful at all.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>     <woods@robohack.ca>
Planix, Inc. <woods@planix.com>;   Secrets of the Weird <woods@weird.com>