RATEHD.DOC 3/5/89 Copyright 1989 ICD, Inc.
This document and the program RATEHD.PRG, may be distributed by any non-
commercial means (may not be sold) as long as the they are distributed
together, and no modifications are made to either, other than language
translation which is allowed. The copyright holder is: ICD, Incorporated,
1220 Rock Street, Rockford, Illinois, USA. 815-968-2228 voice 968-2229 BBS.
RATEHD.PRG was written in Personal Pascal, a product of ICD, Inc.
----------------------------------------------------------------------------
RATEHD was written to show performance characteristics of various hard disk
drives. It was developed at ICD to allow fair comparison when evaluating
hard disk drive mechanisms, embedded SCSI drives and SCSI controllers when used
on the Atari ST. It does not go through a hard disk handler (like ICDBOOT.SYS)
so it is not useful in rating the speed of different handlers. (There are no
significant speed differences in the different handlers anyway. Atari's,
Supra's, and ICDBOOTX.PRG are all the same speed and as fast as anyone can go
on the ST. ICDBOOT.PRG is 2% to 20% slower on writes since it verifies data.)
We were a bit apprehensive about releasing this to the public because of our
fears that people not understand it and will use it for the wrong purpose.
It is intended as a comparison test and should not be used to verify a drive
or controller against printed specifications. Drive manufacturers test their
drives on specialized test equipment and not on an Atari ST. Data rates are
also usually shown as a burst rate and what we are measuring is the sustained
rate which is closer to the top end in real life usage.
Our final decision to release RATEHD was twofold.
1) There is too much misinformation out there when it comes to computers and
hard drives. There needed to be a uniform benchmark test for the Atari ST.
Magazine reviews have typically mislead the public, not maliciously, but
because of the reviewer's ignorance on the subject and lack of scientific
testing methods. Now you have a benchmark; just don't abuse it.
2) When we finished the FA-ST Tape Backup, we realized that a hard drive system
would need a certain amount of finesse to keep up with the streaming tape.
If the tape had to keep backing up and waiting for the hard drive to catch
up, the backup time would become laboriously long and almost not usable. A
data rate of about 300 K/s (kilobytes per second) is required to keep the
tape moving without any stopping and backing up of the tape. As the data
rate gets slower, the tape will begin to backup and restart several times
per partition. A data rate of less than 200 K/s would probably become
unbearable at attempting a backup.
We have tested many drives and controllers at their optimum interleaves and
found that Adaptec MFM controllers (including Atari SH204, Megafile 20) run
about 300 K/s if they are formatted at 1:1 interleave. Adaptec RLL
controllers (including Atari Megafile 30 and 60) run about 315 K/s at 2:1
interleave. Omti MFM controllers move along about 360 K/s using 1:1 with
Omti RLL at about 375 K/s at 1:1. Embedded controller drives are usually
the speedsters with the Seagate ST138N and ST157N cruising at 550 K/s at
1:1. The quickest we have measured was a CDC 94181-702 which flies at
1010 K/s which is probably fast enough to do real time video. It also had
15 ms average access time and had a 600 Megabyte formatted capacity.
THE PROBLEM WITH MOST TEST METHODS
There are many variables in something as complex as a computer with hard disk
drive that can make testing a cumbersome task. The most obvious way to test
hard drive speed, is to copy files from a hard drive to a RAMDISK and then back
to a clean hard drive while timing the process. Although on the surface, a
test like this appears sufficient, this cannot give an accurate picture of what
to expect in performance after using the drive for awhile. Both hard drives
would need to be freshly formatted (no files yet) and with identical partition
sizes. If not freshly formatted, the DOS fat table delays and fragmentation
would enter a significant amount of error into the rating equation. Of course
the same DOS and same computer must also be used.
While this seems like a reasonable test method, it would only show the data
rate (assuming both drives were formatted at their optimal interleave). Step
rate in this type of test would not be a measurable factor as it would later
on when the files on the drives become fragmented. Fragmentation occurs as a
hard drive gets used and files are modified. They increase in size and when
rewritten to the drive, not all the file parts are in consecutive sectors
anymore, probably not even in the same area of the disk. The file eventually
becomes a group of "fragments" and this is where the movement of the head from
track to track (step rate measured as average access time) really becomes an
important speed factor. It would not be an easy task to create two freshly
formatted drives of different manufacture with identical fragmentation on each
drive.
Many of the magazine reviews we have seen give no indication of the test
method. It is almost as if the numbers were arrived at by magic. We now offer
RATEHD as a solution to all magazine reviewers desiring to test hard disk
drives. If you all use this, then the results can be believable.
Since RATEHD does not use the DOS nor the hard disk handler, it is easily
repeatable and almost foolproof; some of the most important factors in any
test method. It will run on any hard drive and does not care about how the
partitions are set up nor does it matter what the file structure is like.
Since RATEHD does *not* ever write to disk, it is totally safe and can be run
at any time.
ABOUT THE PROGRAM
RATEHD begins at SCSI ID# 0, Logical Unit Number 0, and proceeds to check
0-3 LUNs for each SCSI ID going through all SCSI IDs 0-7 except for the special
cases for ID# 6 (reserved for our real time clock) and ID# 7 (where it only
reads LUN 0). When the program finds a hard drive present, it will print the
manufacturers identification under Device Name, and proceed with the tests.
After each drive is tested, it will display the results and search for the next
drive. When finished with all drives, RATEHD will allow you to RETEST or EXIT
back to the desktop. (A screen dump will print the results.)
Don't worry: It is only a read test so no damage will be done to your drive
media.
ABOUT THE TESTS
The Data Rate Test
In this test, one megabyte of data is read (2,000 sectors) in 100 sector
continuous blocks (burst mode). This is timed and the math is done to
calculate the number of K/s (Kilobytes read in one second).
The Average Access Test
The Average Access is based on an assumed 16 megabyte partition beginning
with sector 0 and ending with sector 31,000. The head is told to read a
sector at the innermost cylinder of the partition area (31,000) then to read
one in the outermost cylinder (0). This is repeated 50 times but each time
the sector number is adjusted so that the head will move 1/50 less distance.
If there were 51 tries, the head would end up on the center cylinder of the
partition (15,500), not stepping at all for the inner and outer cylinders.
This movement is timed and the sector read time is removed. An average time
for the head travel is then calculated and shown in ms (milliseconds). Since
this average time also includes the command overhead time, it will be slightly
higher than specifications from the drive manufacturer.
WHAT YOU HAVE LEARNED
There are really only two significant variables concerning performance with a
hard disk. These are Data Rate and Average Access. We are assuming that you
are concerned about speed and have the optimum interleave on your drive.
There is one other bottleneck that does affect hard disk performance. The
Disk Operating System is a major slowdown. To optimize your DOS speed get
TOS 1.4 from Atari as soon as you can. TOS 1.4 is in beta test at this time
so it is not widely available. Once it is, get it; we have seen a tremendous
improvement in both speed and reliability. Thank you Atari!