> > That's an impressive performace hit!
> > 
> > I assume that you directed the output to a file. Both machines ka650?
> > 
> > This is a serious performance degradation, and it _should_ be fairly easy
> > to figure out where the hit took place.
> 
> yeah.. no, the machines werent identical. the slowness was on a KA420, my 
> faster scores were on a KA650. Still, theyre comparable CVAX processors, 
> and a factor of ten speed difference is not explained by the machines. 
> the output was sent to /dev/null in both cases- i dont think that the 
> kernel would be 10x slower in /dev/null, either... 

Looking at the vup chart my MV3100/10 is 2.4, which is roughly 2.5
times faster than the MVII.  The KA650 (MVIII boxes are in the 2.8-3.8
vup range, I am thinking.  I might expect a 50% difference, but not
1000%.  Yikes.

OK, for the sake of discussion, concerning the kernel configgypudding,
what, in NetBSD can be tuned to maximize speed on the config.

Here is the VS3100 config used on the test MV3100/10:

#	$NetBSD: VS3100,v 1.71.2.6 2000/10/04 08:24:27 itojun Exp $
# VS3100 VAX configuration file; all supported devices.
include		"arch/vax/conf/std.vax"
#ident 		"VS3100-$Revision: 1.71.2.6 $"
options 	"VAX410"	# VS 2000
options 	"VAX43"		# VS 3100/76
maxusers	8
options         INET		# IPv4
file-system 	FFS
file-system 	CD9660
config          netbsd root on ? type ?
mainbus0	at root
vsbus0		at mainbus0		# All VAXstations
le0		at vsbus0 csr 0x200e0000 # LANCE ethernet
dz0		at vsbus0 csr 0x200a0000 # DC-367 DZ-11 lookalike ctlr
si0		at vsbus0 csr 0x200c0080 # VS2000/3100 SCSI-ctlr
si1		at vsbus0 csr 0x200c0180 # VS2000/3100 SCSI-ctlr
scsibus*	at si?
sd*		at scsibus? target? lun?
st*		at scsibus? target? lun?
cd*		at scsibus? target? lun?
ch*		at scsibus? target? lun?
ss*		at scsibus? target? lun?
uk*		at scsibus? target? lun?
pseudo-device   loop	1
pseudo-device   pty	64		# pseudo-terminals
pseudo-device	vnd	4


Here is the MVII config:


#	$NetBSD: GENERIC,v 1.71.2.6 2000/10/04 08:24:27 itojun Exp $
# STRIPPED MicroVAX II configuration file.  RDK.
include		"arch/vax/conf/std.vax"
#ident 		"MVII-$Revision: 1.71.2.6 $"
options 	"VAX630"	# MV II
maxusers	8
options         INET		# IPv4
file-system 	FFS
file-system 	CD9660
config          netbsd root on ? type ?
mainbus0	at root
ibus0		at mainbus0		# All MicroVAX
uba0		at ibus0 		# Qbus adapter
uda0		at uba?	csr 0172150	# UDA50/RQDX?
mtc0		at uba? csr 0174500	# Tape MSCP ctlr
qe0		at uba? csr 0174440	# DEQNA/DELQA
dhu0		at uba? csr 0160440	# DHU-11
mscpbus*	at uda?
mscpbus*	at mtc?
ra*		at mscpbus? drive?	# MSCP disk
rx*		at mscpbus? drive?	# MSCP floppy
mt*		at mscpbus? drive?	# MSCP tape
pseudo-device   loop	1
pseudo-device   pty	64		# pseudo-terminals
pseudo-device	vnd	4


There is nothing out of the ordinary in either of them.
They look reasonably plain and essentially the same as
the configs in 4.3BSD.

But, for the sake of discussion, is there anything that
could speed them up?  Is there anything in the other config
files, e.g, std.vax, that could be tuned up some?

A 1000%(?) hit is hurting the throughput, for sure.  I did not
realize some things were getting hit that hard....(:+{{...
I guess it only shows on a  s l o w  class machine.

Thanks

Bob