Subject: Re: MAXPHYS
To: Manuel Bouyer <bouyer@antioche.lip6.fr>
From: Thor Lancelot Simon <tls@rek.tjls.com>
List: port-i386
Date: 02/04/2001 14:26:29
On Sun, Feb 04, 2001 at 02:27:19PM +0100, Manuel Bouyer wrote:
> On Sat, Feb 03, 2001 at 05:57:26PM +0100, wojtek@wojtek.from.pl wrote:
> > could it be changed to /usr/src/sys/arch/i386/include/param.h
> > to larger value (like 256KB or 1MB)
> > 
> > my drive can do 27MB/s with 10ms seek - it could transfer 270kB in one
> > seek time - so 64kB is too small to make clustering work efficiently.
> 
> I'm not sure this is a great idea. You may suffer from other problems,
> especially I/O latency. You'll also use more memory.

It seems like a great idea to me -- if your drive can accept 256K as fast
as it can accept 64K (and most can) there should be little or no substantial
increase in latency.  Since how much memory you use depends on the size of
the buffer cache, not on how large your transfers are, I don't see how that
becomes a problem.

Finally, things like RAIDframe *really* want the old MAXPHYS limit to go
away so that they don't ultimately send 8k transfers to the disks.

> Some drivers will also break (pciide should work but I've never checked - 
> siop will definitively not work without changes).

pciide totally loses with transfers larger than 64k.  I'm surprised that siop
doesn't work -- what limits it to 64k?

Thor