Subject: Re: UBC status
To: Manuel Bouyer <>
From: Giles Lean <>
List: tech-kern
Date: 09/26/1999 09:24:28
On Sat, 25 Sep 1999 17:12:50 +0200  Manuel Bouyer wrote:

> On Sat, Sep 25, 1999 at 01:26:29PM +1000, matthew green wrote:
> > 
> >    - add limits on how much of physmem is used for cached file data?
> > 
> > 
> > there should definately be limits on this...
> For unmodified data I'm not sure there should be one (why ?) but for
> dirty buffer there should. It's should'nt be possible for someone to eat all
> the RAM just by writing a large file.

Some commercial Unix variants (which shall go nameless to protect the
guilty) have found the VM system getting very unhappy when the buffer
cache gets to several GB in large systems, leading to apparent hangs,
support calls, etc etc.

Now, NetBSD has a bright shiny new VM system and may not face the same
scaling problems, but I'd be somewhat wary!

A second reason to limit the amount of memory used for the buffer
cache is for more consistent application response -- i.e. to ensure
that there is always memory free for applications, even if it would be
more efficient in total throughput to allocate this memory temporarily
for buffer cache.