On Wed, Jun 19, 2002 at 04:51:56PM -0400, Greg A. Woods wrote:
> [[ note redirect to tech-perform ]
> 
> [ On Wednesday, June 19, 2002 at 21:56:01 (+0200), Wojciech Puchar wrote: ]
> > Subject: Re: NetBSD 1.6 Release Schedule
> >
> > > > i hope UBC will be really fixed. present behavior is inacceptable.
> > >
> > > UBC seems to be working fine. What do you think is wrong with it? Or
> > 
> > not only think but see and already did send-pr.
> > 
> > on heavy disk usage whole machine slows down incredibly and there are 5-10
> > seconds reaction in X programs. using navigator etc. is almost impossible.
> 
> That's because, I believe, you have not yet properly tuned your system
> (i.e. your kernel and perhaps also some of the "tunable" resources) for
> your most common application mix.

Really, I have to agree with Wojciech, a hint or howto pointing out all
the knobs, as well as reasonable values for those, would be rather helpful
for the gros of the NetBSD users imho. That is not to say that the defaults
are unusable, just saying that it would make research on how to tune your
system to your needs easier.

I would be happy to read (sorry, not knowledgable enough to _write_) such
a guide.

Btw, I personally sacrifice quite much of my RAM for file data, as well as
I hard-wire certain min values for other things, and I have to say that I'm
quite content with my settings. Otoh, I haven't played enough with those
(those were inspired by some hints from chuck silvers back in the beginnings
of the merge of UBC into -current I believe) values to tell if it could get
better, but from what Wojciech sees, I believe it can get way worse...
(And no, this is not backed up with vmstat or anything data, that's just
"how it feels". Yes, that's not a good measurement, but otoh, if it felt
bad, I wouldn't have to measure much to realize that it's not too good :))

[for those interested: anonmin/max 20/80 execmin/max 30/30 filemin/max 20/50,
yes, that anonmax prolly should be 50, seeing execmin+filemin==50 % :)]

-Martin