On Fri, 27 Aug 2004, Johnny Billquist wrote:
> > Wouter Klouwen <dublet@dublet.org> typed:
> > : I'm running a -current kernel on my laptop, which has 256MB of internal
> > : memory, and a quite slow HDD. The filecache is usually ~170MB, this is
> > : really way too much,
>
> I can't speak for Wouter, but in my case, the system *is* doing things
> wrong.
> It's not a laptop, but a stationary machine, with rather fast SCSI. And it
> acts both as my NFS server, and as my interfactive plaything.
> The current "hit" for me came because I have another machine that was
> building X, which had xsrc nfs-mounted from this server. At the same time
> I was reading mail with netscape. The file cache grabs a lot of memory,
> and forces a large number of pagefaults in Netscape. The result is that a
> number of other programs, also needing a bit more memory also get hit.
> Thus, my machine got about 90% idle, a load of about 4, and about 200
> pageing in/out ops per second.

Raise your {exec,anon}{min,max}, lower your filemax. This has the effect
that when there is competition for memory, file cache will end up with
less, and more of your executables and their data stays in memory.

> It's *obviously* not a good balance in there, for atleast my useage.

Well, it can be hardly expected that the defaults NetBSD has will be
right for everyone. And it seems that your usage is, at least slightly,
uncommon.

> I *don't* have too much ram, I obviously have too little, and at that
> situation, things aren't really fun anymore. NetBSD used to be much faster
> in the past for me. (And no, the answer isn't always "buy more memory".)

Perhaps the problem is that with the advances NetBSD has made, you
need to do more configuration to find optimal settings. But until there
is some intelligent routine that tunes the vm parameters automatically
as load shifts, there's not much to be done about it.