evans@syd.dit.CSIRO.AU (Bruce.Evans) (10/22/90)
On a new 33MHz 386, I found that the 1.2M TEAC floppy drive performed much worse than the 1.2M Fujitsu drive on an old 20MHz 386. It takes about 50 sec to read all the files off a nearly full disk under DOS on the new system, versus 30 sec on the old system. Swapping the drives proves that the new drive is at fault. Usually I use these systems under Minix, where I have fine-tuned the drivers for optimal performance - at least with the old drive. The drivers assume that the head can be stepped across one track in less time than it takes for the index-hole gap to rotate under that head. (This is used for steps at the index-hole gap. An extra sector-time is allowed for steps elsewhere. This gives better performance than the usual BIOS drivers.) The new drive apparently does not meet the assumed spec, so it ends up wasting one revolution in every 3 (2 revs to read 2 sides, then 1 rev wasted stepping) for a bit less than 67% efficiency. The old drive gives 95% efficiency. The step rates of the drives are almost exactly the same, at least for 80- track seeks. Both take about 4 msec per step. So what is wrong here? 1. Some floppy drives are no good? Many? 1a. Certain manufacturers drives are no good? 2. Some drives are better than may be expected? 2a. ... 3. Some drivers assume too much of the hardware? 4. Few drivers handle out-of-spec hardware well? It will be easy to change the timing in the Minix drivers to get close to the old efficiency. Unfortunately, this will cost 6% efficiency (one sector every 15) for drives like the old one. -- Bruce Evans evans@syd.dit.csiro.au