On Wednesday 03 September 2008 13:53:09 Matthias Bethke wrote: > Hi Alan, > > on Wed, Sep 03, 2008 at 08:57:42AM +0200, you wrote: > > These days the entire concept of a "cylinder" is a mere abstraction to > > make tools like fdisk work in a sane manner. > > Of course not. The disk is physically organized in cylinders, that's the > structure dictated by the mechanical design. That a disk controller is > theoretically free to map cylinders and sectors to whereever it pleases > doesn't mean that there wasn't a direct relationship between cylinder > number and physical location on the platter in the vast majority of > non-broken (i.e. cylinder-remapped) disks. With many HD tests in > magazines you get a cylinder-vs.-transfer-rate plot and it still mostly > matches the old rule. I suppose not even firmware hackers are really > eager to make things more complicated than absolutely necessary :)
I disagree on the detail. It's true enough that the data is written in concentric circular patterns following the apparent path the head follows. However, outer tracks contain more sectors than inner tracks. fdisk reports the same size per cyclinder regardless of it's number. So what fdisk calls a cylinder cannot possibly map exactly to a physical circular track. Putting this in firmware is no big deal. If you have 100,000 sectors and 10,000 tracks, then sectors 0-9 are cylinder 0, sectors 10-19 make cylinder 1 etc etc. A simple calculation translates between the two plus a lookup table to account for marked badblocks. This is grossly over-simplified but you get the idea. However, it does make the most sense to keep fdisk's cylinders in some sort of sequential order, so low numbered cylinders will in all probability end up near one edge and high numbered cylinders at the other edge. I strongly suspect that you know this also, and we actually do have the same understanding of how it works :-) -- alan dot mckinnon at gmail dot com
