[ On Wednesday, May 7, 2003 at 08:44:12 (+1000), Paul Ripke wrote: ]
> Subject: Re: some performance diff with getc()/putc() between FreeBSD and NetBSD?
>
> The only other thing that comes to mind is allocation policies and
> filesystem layout - allocation groups, maxbpg, etc. A smallish 3 GB file
> may be spread thinner over the disk, and disksort can't get much of a win,
> 8 GB file is spread a little thicker and disksort can get more benefits.

It would indeed really help to have instrumentation in all the block
device drivers to allow recording of all the transaction addresses as
well as their sizes (and of course direction).  Then I could then plot
the block addresses accessed across time.  So far I've only encountered
ways of doing this using independent bus analyzers, but I've been trying
to think of ways to do this efficiently in the kernel and without
causing more disk I/Os, maybe by spewing them out as UDP packets,
perhaps even with multiple records per packet.....

I suppose one could adjust the newfs parameters in various ways to
try to perturb the results and see if they shift enough to maybe account
for this hypothesis.

-- 
								Greg A. Woods

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