It is of course a natural extension to consider whether range reads should also be supported by a disk. Range reads would be useful in a number of contexts: for example, to pick the rotationally-closest block replica [16,38], or to implement efficient ``dummy reads'' in semi-preemptible I/O [9].
However, introducing range reads, in particular for improving rotational costs on reads, requires an expanded interface and implementation from the disk. For example, for a block to be replicated properly to reduce rotational costs, it should be written to opposite sides of a track. Thus, the disk should likely support a replica-creating write which tries to position the blocks properly for later reads. In addition, file systems would have to be substantially modified to support tracking of blocks and their copies, a feature which only a few recent file systems support [31]. Given these and other nuances, we leave range reads for future work.
0