Subject: Re: large disk preformance (interactive lags in 1.5.1-BETA)
To: Dwight Tuinstra <firstname.lastname@example.org>
From: Jaromír <email@example.com>
Date: 04/09/2001 21:12:05
If you have plenty of memory and are using 1.5, you may try to bump
BUFCACHE or BUGPAGES (see options(4)) or/and use MFS on /tmp.
NetBSD -current has integrated UBC, so it uses all available machine
memory as I/O buffers automatically. Thus, the reponsiveness
should be (much) better on -current for your scenario (plenty of RAM,
Dwight Tuinstra wrote:
> Tyrel Beede wrote:
> > I recently replaced my 4.5 GB UW SCSI @ 7200 rpm drive with an IDE 30 GB
> > @ 5400 rpm. The switch wasn't optional and I've come to notice that the
> > new drive lags with interactive preformance every now and then. For
> > example I'll be running a vi session and things will just kinda bog
> > down, it seems to occur at random points( some of the time just after a
> > reboot and others after the machine has been up for a few days ). The
> > lag is kinda like working with a vi session over a slow telnet link.
> > I'll type on the keyboard and a few seconds later the commands will show
> > up. I've got 96 megs of ram in the machine and I don't even have X
> > installed so the machine isn't really being hit hard for memory. This
> > problem isn't really that big because the lag only happens every now and
> > then... maybe once a day. But it never happend on my SCSI drive running
> > the same kernel. Anybody have any tuning ideas?
> I've had a similar problem with response lag lately. The
> system is a fresh install from snapshot of 1.5.1-BETA. The
> problem isn't just vi, though. I sometimes notice it just
> when doing an `ls':
> 1) login to several virtual ttys (not even in X, just on console)
> 2) start a disk-intensive job running on one or more ttys
> (eg, untar pkgsrc on one and rebuild kernel on another)
> and let it run for a while.
> 3) switch to a free tty and type `ls'. 1-3 second lag.
> 4) typing `ls' again immediately gets a quick response.
> The disk is an older 5400 rpm IDE model, PIO 4. There's also
> another, even older 5400 rpm drive on the IDE cable, but it's
> not even mounted (but I think I saw the behavior even when it
> was completely disconnected). Unfortunately, it'll be at least
> a week or two before the HDs are upgraded.
> >From the behavior, I'd guess its something to do with buffer/
> cache behavior not being tuned. The system I'm seeing this on
> has plenty of memory (256Mb) and speed (1Ghz Athlon T-bird on
> an IWill KK266 mobo (latest VIA Athlon chipset) ).
> If more details are wanted, I can supply them (the machine's at
> home, modem not yet working right either, so this is written
> from memory while at work). Let me know what data you'd like.
> --Dwight Tuinstra
Jaromir Dolecek <jdolecek@NetBSD.org> http://www.ics.muni.cz/~dolecek/
NetBSD - just plain best OS! -=<>=- Got spare MCA cards or docs? Hand me them!