On Mon, 13 May 2002, Thor Lancelot Simon wrote:
> If we were going to do *that*, it would make sense to rip out all the
> rotational placement and other obsolete complexity at the same time.  But
> once you start to go there, it starts to look an *awful* lot like it might
> just make sense to build a simple filesystem that mostly trusts the disk to
> place blocks.

that stuff is definitely not all obsolete - or at least, that _kind_ of 
stuff is definitely obsolete.. it still pays to know the geometry of a 
disk and use that knowledge. 
for example, one of the very cheap things with a pretty good win is to 
avoid allocating blocks which span track boundaries, as a quick way to 
(even if at the cost of some space) force track-aligned accesses.. the 
result is noticeable. there are various other tricks and extra things one 
can do if one knows where the data is on the platter (and even more if 
one can figure out where the head is)

isildur