Hard Disk dd benchmarking

[Jeff Liebermann]

I was trying to get some perfomance figures of various OS/machine/drive
combinations and discovered some erratic results.  The original testing
was to determine how detrimental sector translation was to drive 
performance.  A simple dd from the raw device should yield a good
approximation of the sequential read performance without involving
the kernel buffering (NBUF).  The blocked read performance was thrown
in just for entertainment value (leaving 100k-600kBytes in buffers).
The results were wildly distributed.  Question:
1.  Is the test valid?
2.  Why the 50:1 (best/worst) variation in speed?
3.  Am I doing something wrong?

Hard Disk Benchmarks.    J. Liebermann  
Thu Aug  2 19:25:55 PDT 1990
===========================================================================
Copies exactly 10Mbytes to /dev/null.
Must be root to read /dev/root.

block:	time dd if=/dev/root  of=/dev/null bs=64k count=160
raw:	time dd if=/dev/rroot of=/dev/null bs=64k count=160
===========================================================================
Size of root partition-\	/----------------------	Controller
Cache Size --------\   |	|	/--------------	Flavour
RAM -----------\   |   |	|	|	/------	Drive
CPU/CLOCK --\  |   |   |	|	|	|	OS Version --\
	    |  |   |   |	|	|	|		     |
comix	386/25 4Mb 0k 125Mb	ACB1542A SCSI	Maxtor 4380S	Unix 3.2.0
	block	real=34.4	sys=12.7	KBytes/sec=344
	raw	real=56.0	sys=6.4		KBytes/sec=178
adcom	386/25 4Mb 0k 80Mb	DTC6280A ESDI	Maxtor 4380E	Xenix 2.3.1
	block	real=181	sys=6.8		KBytes/sec=55
	raw	real=97		sys=3.5		KBytes/sec=103
bucci	386/25 4Mb 0k 65Mb	WD1006 1:1 MFM	MinScr 3085	Xenix 2.3.1
	block	real=180	sys=6.4		KBytes/sec=55
	raw	real=105	sys=3.8		KBytes/sec=95
aton	386/25 12Mb 16k 100Mb	ACB1542A SCSI	MinScr 9380S	Unix 3.2.0  (1)
	block	real=32.6	sys=8.0		KBytes/sec=306
	raw	real=53.8	sys=3.8		KBytes/sec=187
orion	386/25 8Mb 128k	90Mb	WD1007A	ESDI	Wren V 380mb	Xenix 2.3.2
	block	real=27.7	sys=5.1		KBytes/sec=361
	raw	real=34.6	sys=2.4		KBytes/sec=289
redwood 386/25 5Mb 64k 35Mb	WD1006 1:1 MFM	MinScr 3085	Xenix 2.3.3
	block	real=27.0	sys=5.6		KBytes/sec=370
	raw	real=25.6	sys=2.6		KBytes/sec=390
redwood	386/20 4Mb 0k 50Mb	Ultrastore ESDI	MinScr 3180E	Xenix 2.3.3 (2)
	block	real=18.4	sys=13.9	KBytes/sec=543
	raw	real=17.0	sys=3.9		KBytes/sec=588
srccom	386/25 8Mb 64k 200Mb	ACB1422B-15ESDI	MinScr 9780	Xenix 2.3.2 (3)
	block	real=81		sys=8.4		KBytes/sec=123
	raw	real=34.6	sys=2.5		KBytes/sec=289
dsch01	386/20 4Mb 0k 80Mb	Ultrastore ESDI	MiniScr 9380E	Xenix 2.3.2
	block	real=17.0	sys=13.4	KBytes/sec=588
	raw	real=16.0	sys=3.3		KBytes/sec=625
grumbly	386/20 8Mb 0k 62Mb	WD1006 1:1 MFM	MicroSci HH1090 Unix 3.2.1
	block	real=24.9	sys=9.4		KBytes/sec=402
	raw	real=24.3	sys=1.4		KBytes/sec=412
vdx	386/25 8mb 0k ?Mb	WD1007A ESDI	Maxtor 4380E & CDC Wren V
	block	real=18.4	sys=17.5	KBytes/sec=550	ISC 2.2
	raw	real=14.2	sys=7.0		KBytes/sec=704
	(2nd controller)	ACB1542B	CDC 94171-344 & ST4702N
	block	real=14.7	sys=14.2	KBytes/sec=680
	raw	real=11.0	sys=0.7		KBytes/sec=910

(1) Using Storage Dimension's SCSI driver.
(2) Ultrastor has 32K on board cache.
(3) Lookahead disabled on controller.
(4) Dos partition on aton only.  Others are all Unix or Xenix.
(5) vdx has 2 controllers and 4 drives.


-- 
# Jeff Liebermann   Box 272   1540 Jackson Ave   Ben Lomond  CA  95005
# (408)336-2558 voice  (408)429-0483 digital pager      CIS:73557,2074 
# PC REPAIR & RF DESIGN.   uunet!comix!jeffl   pyramid!vdx!comix!jeffl
# universe!milky_way!solar_system!earth!na!us!uunet!comix!jeffl