aha8scsi plus SCSI performance

QNX Newsgroups (Archives) qdn.public.qnx4
1

Hello,

I just did some tests that don’t make any sense to me.

I tried out a new adaptec 29160 controller
( I believe it is a regular 29160 not 29160N -
it has 2 68 pin int. connectors, 1 50 pin int. and 1 68 pin ext. )

I used a program which writes, reads, and verifies a number of files.

I used the 4.25E aha8scsi driver with the 29160.
I also used the 4.25E aha8scsi driver with a 2940uw controller.
I used the aha7scsi driver with the 2940uw.
I used the sym8scsi driver with a NCR scsi controller.

I used the same hard drive for all the tests - a seagate ST39236lw
7200rpm.

all the controllers performed about the same
(which for only 1 drive connected to the controller seems right)

Here are some of my results…
parameters (-s {size in bytes}, -f {# files} )

-s1000 -f 250 - time 10.35 secs
-s50000 -f100 - time 6.85 secs
-s5000000 -f10 - time 5.5 secs
-s18500000 -f5 - time 12.5 secs

Does this seem right ???

What I really don’t get is this…
I tried an IDE drive - Seagate ST 315320A 7200rpm

I thought it would be much slower.
-s1000 -f 250 - time 0.95 secs
-s50000 -f100 - time 0.65 secs
-s5000000 -f10 - time 2.5 secs
-s18500000 -f5 - time 27.0 secs

This doesn’t make any sense to me.

The computer was a PIII-933 with 256M RAM.
Fsys running with 40% cache

-Glenn Sherman

2

Glenn Sherman wrote:

Hello,

I just did some tests that don’t make any sense to me.

I tried out a new adaptec 29160 controller
( I believe it is a regular 29160 not 29160N -
it has 2 68 pin int. connectors, 1 50 pin int. and 1 68 pin ext. )

I used a program which writes, reads, and verifies a number of files.

What does verify do?
Does the test write, read and verify one file, then go on to the next
file or does it write all files, read all, etc.?

I used the 4.25E aha8scsi driver with the 29160.
I also used the 4.25E aha8scsi driver with a 2940uw controller.
I used the aha7scsi driver with the 2940uw.
I used the sym8scsi driver with a NCR scsi controller.

I used the same hard drive for all the tests - a seagate ST39236lw
7200rpm.

all the controllers performed about the same
(which for only 1 drive connected to the controller seems right)

Here are some of my results…
parameters (-s {size in bytes}, -f {# files} )

-s1000 -f 250 - time 10.35 secs
-s50000 -f100 - time 6.85 secs
-s5000000 -f10 - time 5.5 secs
-s18500000 -f5 - time 12.5 secs

Does this seem right ???
Hard to say without more info. There is Fsys overhead to create

a file; the difference between long serial reads and writes vs
short random access is very large; etc.

What I really don’t get is this…
I tried an IDE drive - Seagate ST 315320A 7200rpm

I thought it would be much slower.
-s1000 -f 250 - time 0.95 secs
-s50000 -f100 - time 0.65 secs
-s5000000 -f10 - time 2.5 secs
-s18500000 -f5 - time 27.0 secs

This doesn’t make any sense to me.

Depends. Did you reboot before running this? How much head seeking
is there on the IDE vs SCSI, i.e. how much does the head have to move
to go between directories, .bitmap(if you are creating or growing the
files) .inodes, and file extents?

The computer was a PIII-933 with 256M RAM.
Fsys running with 40% cache

That’s enough for everything to be in cache if there is nothing else
using significant memory. On the other hand, I have heard that Fsys
tries to guess if you are doing serial reads/writes and won’t cache
those.

-Glenn Sherman