Subject: Re: BUFFERCACHE, PR 1903
To: Chris G Demetriou <Chris_G_Demetriou@ux2.sp.cs.cmu.edu>
From: Michael L. VanLoon -- HeadCandy.com <michaelv@MindBender.serv.net>
List: current-users
Date: 09/14/1996 09:29:02
>> >	(4) it's annoying to no longer be able to have the "N percent
>> >	    of the first M megs, P percent of the rest" function,
>> >	    to better support small-memory systems.

>> There's no reason you couldn't default back to this if BUFCACHE (or
>> whatever) wasn't defined.  Say, for GENERIC kernels.

>So, that means that the diffs are machine-dependent, and there's a
>question about whether or not what the patch does is 'right'...  that
>means that somebody's gotta take the time to figure out 'the right
>thing.'
[...]
>If you're trying to use a consistent kernel throughout a given
>environment, and you claim that you don't have a fixed amount of
>memory in each machine (so don't want to use NBUF/BUFPAGES), who's to
>say that you wouldn't hit the "small memory" case, or that having a
>more carefully tunable amount of memory wouldn't be generally useful?
>(In other words, i'm still not convinced that this should be an
>option at all, and that if it should be it's "good enough" as is.  8-)

I think you're looking at it from the wrong direction.  Assume that
you're a sysadmin responsible for a site with several machines, and
this option is available.  You build a "standard" kernel for use on
the majority of the "general-purpose" workstations.  You don't use the
BUFCACHE options, but let the default case happen.

You also build a "standard" kernel for use on servers.  You do use the
BUFCACHE option because these servers don't run a lot of big
processes, but hit the disk a lot.  This is the exception to the
general case.  But it's a lot easier to do if there's a one-line
option you can add to the kernel config file.

I'm taking this experience from actual experience helping run things
at Iowa State.  But there, we even set the BUFCACHE percentage for the
workstations, if I remember right.  I believe it was either 10% or 20%
for the general case.

-----------------------------------------------------------------------------
  Michael L. VanLoon                           michaelv@MindBender.serv.net
        --<  Free your mind and your machine -- NetBSD free un*x  >--
    NetBSD working ports: 386+PC, Mac 68k, Amiga, Atari 68k, HP300, Sun3,
        Sun4/4c/4m, DEC MIPS, DEC Alpha, PC532, VAX, MVME68k, arm32...
    NetBSD ports in progress: PICA, others...
-----------------------------------------------------------------------------