Subject: Re: diffs for UVM/UBC improvements available
To: Bill Sommerfeld <>
From: Chuck Silvers <>
List: tech-kern
Date: 05/22/2001 08:26:20
On Tue, May 22, 2001 at 08:50:15AM -0400, Bill Sommerfeld wrote:
> My overall reaction is that this is a pile of largely independant
> changes which might be better reviewed, tested, and absorbed as
> several independant sets of commits.  (e.g., one blob of VOP_ISLOCKED
> nuking, ...).  Yes, this is more work, but there are 3/4 meg of diffs
> to review here...

I can't disagree with this, but I don't think it's really that much harder
to look at now.  I just read through the whole diff last night to write up
the summary, and it wasn't hard to tell which aspect different parts were for.
let's try it this way and see how it goes.

> >    (I also got rid of v_lease on the principle that it's an optimization for
> >    an almost-unused feature.)
> Why?  I actually use this stuff.  How much slower is NQNFS under heavy
> use since making this change?  This strikes me as "going a step too
> far".

I have no idea what the performance effect of this is.
what test would you be interested in, specifically?

> > as I was working on this stuff, I didn't really expect that it would have
> > much affect on a more complex task like building a NetBSD snapshot.
> > to my surprise, it does!  on my 400MHz pc, I see this improvement in
> > "make build" performance:
> > 
> > -current:
> > 4852.751u 779.388s 2:50:07.17 55.1%     0+0k 153979+322668io 42565pf+0w
> > 
> > with these diffs:
> > 4855.911u 883.693s 2:29:39.24 63.9%     0+0k 16927+58502io 15800pf+0w
> oddly, cpu usage increases while walk-clock time decreases (and i/o
> decreases dramatically).  do you have any understanding of why this is
> the case?  

my theory is that cpu usage goes up because of increase cache and TLB
footprint.  I don't know why it's faster, I was surprised to find that.