Wierd IDE/Triton behavior

Benson L Chow (bc3c+@andrew.cmu.edu)
Thu, 9 Jan 1997 00:17:35 -0500 (EST)

This probably isn't exactly a kernel problem, since I had drives that
work fin ewith the current linux IDE/Triton driver. However, I have a
IBM DAQA32160 drive that is getting strange timing results with
hdparm(8). In MSDOS/Win95 I get around the expected 6MB/sec transfer
rate. However, under Linux I can get anywhere from 1.8MB/sec to
3.5MB/sec. with hdparm. Dumping with dd direct from disk to /dev/null
gives similar results. What's strange is the other drives I've had
(that supports mode3 only) can get consistent 3.1MB/sec though it's
clearly a older, slower drive (Maxtor 7850). This clearly points to a
possible problem with the drive itself but apparently other IBM
DAQA31260's are also implemented this way. I'm just wondering if anyone
has seen something like this happen on their drives.

For reference:
IBM DAQA 32160 2.1GB (5400rpm) PIO Mode4 Harddrive. Win95 BMIDE- 6 MB/sec
Maxtor 7850 850M (3400rpm) PIO Mode3 Harddrive. Win95 BMIDE- 3.2MB/sec
(Program used: Adaptec EZ-SCSI Disk Benchmark for sequential disk access)

Under linux: IBM DAQA gets random 1.6MB/sec to 6 MB/sec, average 2 MB/sec
Maxtor gets about the same as windoze... 3.1 MB/sec consistently

These were tested on unloaded computers with no significant background
tasks and repeated several times.

The IBM drive was used on a Triton 82371SB Triton HX and a 82371FB Orion
KX board with similar results (The FB was a little slower - but not
significatly).

The Maxtor drive was tested on the Triton 82371SB chip only (before it died)

Thanks for any suggestions, and/or anyone who gets similar results.
-bc