Re:The Hairy State of Linux Filesystems
by jacquesm on Thursday February 12, 2009 @12:05AM
(#478056)
Not 'totally true', but it was a lot more true in the past than it is today, logical block addressing on harddrives (especially PC drives, scsi and other non-consumer targeted drive architectures much less so) is a relatively new thing.
In the past the host did a lot more of the work that has now been offloaded to the drives, since IDE basically the whole controller moved off host onto the drive. The reasons are simple, tighter integration and lower costs as well as a uniform interface to the drive, no matter what goes on physically inside it. Nobody really cares! (well, drive manufacturers obviously do, but consumers will treat it like a black box, it either works or it doesn't, no much point in figuring out how it works).
So, flash in this sense is roughly where harddrives were in the times of the early winchester, you present a pretty rough device to the outside and the host has to know a lot about how it works to make it function.
SSDs are already changing this, the controller is built in now and the interface presented is very close to the one used by a normal harddrive, in fact most are indistinguishable. Even most usb sticks and the larger cards do a good job of hiding the gory details.
That full rotation business is not quite the whole story by the way, plenty of drives nowadays do not need to see the 'whole' track in order to do a simple read or write of a sector, they can optimize for that to the point that the sector will be just about ready to be read from or written to by the time the drive head arrives above the track.
Most drives try to avoid that situation as long as they can though, and most file systems do too by picking a block size that is a fairly large multiple of a sector.
In the past the host did a lot more of the work that has now been offloaded to the drives, since IDE basically the whole controller moved off host onto the drive. The reasons are simple, tighter integration and lower costs as well as a uniform interface to the drive, no matter what goes on physically inside it. Nobody really cares! (well, drive manufacturers obviously do, but consumers will treat it like a black box, it either works or it doesn't, no much point in figuring out how it works).
So, flash in this sense is roughly where harddrives were in the times of the early winchester, you present a pretty rough device to the outside and the host has to know a lot about how it works to make it function.
SSDs are already changing this, the controller is built in now and the interface presented is very close to the one used by a normal harddrive, in fact most are indistinguishable. Even most usb sticks and the larger cards do a good job of hiding the gory details.
That full rotation business is not quite the whole story by the way, plenty of drives nowadays do not need to see the 'whole' track in order to do a simple read or write of a sector, they can optimize for that to the point that the sector will be just about ready to be read from or written to by the time the drive head arrives above the track.
Most drives try to avoid that situation as long as they can though, and most file systems do too by picking a block size that is a fairly large multiple of a sector.