Subject: Re: bufpages calculation
To: Chuck McManis <>
From: David Brownlee <>
List: tech-kern
Date: 09/27/2000 07:39:03
On Tue, 26 Sep 2000, Chuck McManis wrote:

> >On Mon, 25 Sep 2000, Bill Sommerfeld wrote:
> > > ... back in the days where 4MB of memory was the norm...
> At 07:11 PM 9/26/00 +0100, David Brownlee wrote:
>  > But kernels were much smaller then.
> Hello tech-kern,
> Not to pick on Bill or David here, but one of the nice things about NetBSD 
> (vs everything else out there) is that it is wonderfully cross platform. 
> Hence I get a bit worried when systems with 4MB are looked upon fondly as 
> ancient relics. When I contemplate deploying systems _today_ that have 4MB 
> or less of main memory. I don't like to hear, "A 2MB system is too small to 
> support UNIX." when certainly versions of UNIX have and do run on such 
> systems (even PDP-11's for heavens sake!)
> This is a very distinguishing factor that certainly the Linux community has 
> embraced (only used as an example of another group where cross platform is 
> a stated goal) and we should keep it in mind as well. Sure 64MB Pentium 
> systems are about $30 at GoodWill but the world isn't defined by such things.

	You clipped the part where I mentioned I was trying to help the
	low memory case :) If you think 4MB on a vax is low, try it on
	an arm26 box - RISC instruction set and a 32K hardware pagesize.

	We are still very much interested in supporting low memory
	machines - Frank has just modified the install setup to help
	4MB i386 boxes which should be applicable to other architectures,
	Ben has added an NFS_V2_ONLY option to exclude NFS3 and NQNFS
	code, and Jason added VNODE_OP_NOINLINE to avoid inlining vnode op
	Howver its inevitable that NetBSD kernels are larger than BSD4.X -
	there is a lot more functionality and hardware support knocking
	around in there - so even using BSD4.X rules for bufpages
	calculation we should be allowing fewer buffers on low memory

			       -- A pmap for every occasion --