On Thu, Sep 06, 2001 at 01:08:05AM -0500, Bruce Anderson wrote:

> >Having other IO come to a grinding halt while running newfs is not
> >exactly desired :-)
> 
> That is from running newfs on the same spindle as / .
> ( Your using a IDE drive, right?)

In my current situation, / is on sd0, and I am newfs'ing sd1.

i/o to sd1 is almost impossible while running newfs, and certainly too
slow to be useful.

i/o to sd0 is acceptable, but slower than it should be. I'm not flooding
the SCSI bus, and i/o to that disk is nil.

Under Solaris on the same hardware, running newfs on a partition and
doing i/o to another partition on the same disk is acceptable, and i/o
to another disk is normal (which supports the fact that I'm not flooding
the SCSI bus).

Perhaps there is an issue with scheduling of i/o calls to the same
device?

I've seen these results using IDE disks on i386, and SCSI on sparc, so
it does not appear to be specific to a particular bus, though I
understand why it might be more of a problem on an IDE bus.

> does nice -n 20 newfs help?

on 1.5.1, at least,

# nice -n 20 newfs /dev/sd1a
nice: Badly formed number.

The man page indicates this is the correct syntax. This obviously a
side issue :-)

Since this is pretty much i/o bound, I don't imagine this is part of
the problem. The only difference it might make is that newfs gets _no_
CPU time at all since other processes want it.. which isn't much
better :)

> To see what is going on run systat -w 1 vm in one window
> then start newfs.

As expected, i/o on sd1 is 100% during newfs, and CPU usage is
marginally higher than normal.

g.