On 02/08/05 11:24:17, anahata wrote: > Multiple platters no problem: for performance reasons sequential sectors > are always read from all platters in one head position before moving the > head. I don't suppose anyone knows if the platters are read in parallel? Assuming sequential reads, the obvious way to work would be to read all the sectors of the first track of the first platter, then all the sectors of the first track of the underside of that platter, then the same for the next platter etc. until the last sector in that "cylinder" was read and then move the head to the next track. Potentially though, much greater throughput could be had for the same rotational speed by instead of switching head and therefore which platter is being read, having a head applifier/decoder for every head so as the disk spins all platters are read in parallel. That way sector 0 would be the first sector of the first track on the first platter, sector 1 would be the sector on the underside of that platter also at track 0 and in the same rotational position etc. With perphaps four platters, i.e. eight sides, sector 9 would be the second sector of the first track on the first platter and so on. > The mapping isn't plain linear, but it will follow a reasonably smooth > head movement or disk cache "elevator" algorithms won't work optimally, > nor will long sequential transfers. Modern disks are zoned, so there are more sectors per track near the edge, but I would still expect all the sectors in any cyclinder closer to the middle of the disk to have higher logical addresses (LBA or SCSI) than those in any track further to the outside. The only reason for departing from this I can think of is if the drive firmware detects a defect in the surface and relocates sectors to a spare track. > Yes, so maybe I was wrong and the low numbers are at the edge and the > high numbers in the middle. I think that's normally the case though I guess there is nothing to stop any particlar manufacturer from doing it the other way round. Steve.
