On Mon, 2006-04-24 at 13:10 +0100, James Wilkinson wrote: > Lyvim Xaphir wrote: > > A drive's sectors actually start out on the rim of the drive. In that > > area, you have more sectors per track (outer rim) than as you go closer > > towards the spindle. At the spindle you have the lowest sectors per > > track. Both the data rate and the sectors per track are greatest at the > > rim of the hard drive; which is where the boot sector is too btw. The > > heads can stay on one track and transfer more contiguous data per > > cylinder on cylinders at the rim than they can close to the spindle. > > This is true, but you also need to take into account how long it takes > for the heads to get to the right bit of the disk. > > When a hard drive is asked to read (or write) data, it has to move the > read heads to the right track. Then it has to wait for the right bit of > data to come around. Thanks for making my point. This is exactly one of the reasons why data transfer is superior at the start of the drive than at the end (spindle). In modern drives the sectors per track are greater at the start than at the end of the drive. Therefore heads can stay in place and transfer more data without moving at the rim of the platters; especially if it is contiguous after a new install within the context of the partitioning scheme I posted. The cylinders at the rim allow more data transfer with less head movement, which adds to better overall data transfer. *Especially* if your partitioning is set up to take advantage of this. > > For short reads and writes (a few hundred K or less), the time for the > head to get into the right place is going to be a lot more than the time > for the head to actually read the data. Thanks again for the support; that's exactly why the outer platter is more valuable than the inner platter. This begs the reason for partitioning as described, which is to relegate 'data' (in this case binaries) to regions of similar function while maintaining performance prioritization. During the boot process, after the root pivot from the initrd, the primary accesses are /boot and /root. (which is exactly why I grouped them together. Plus the room they take up is marginal in comparison with the rest of the drive) The head movements you speak of are minimal in the partition layout I described during this time period. Same thing for /swap and /user after the system gets up and running. When a system is first installed, all data is basically contiguous. So in that scenario head movement is even further minimized in the context of the partitioning plan I outlined. > > So for small accesses, you do better by locating the data in the middle > of the drive somewhere, so the head hasn't got quite so far to travel > and can get there faster. The heads have the least travel at the rim while recovering data and this becomes progressively less true as you progress towards the spindle. If you are saying that the hard drive is most efficient at the middle, this is wrong. If you are saying that data should be stored in the middle, this is academic because the partitioning scheme I posted does that anyhow. > > It's not unknown for administrators to limit the amount of data that > they will store on a disk so the heads don't have to travel too far... Yes, that's one purpose of partitioning. LX -- °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°° When the ax entered the forest, the trees said, "The handle is one of us!" -- Turkish proverb Registered Linux User #268899 http://counter.li.org/ °°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°°
