Our study of disk scheduling has revealed a number of interesting results. First, the overhead of positioning time cannot be avoided with traditional SPTF scheduling alone; even with multiple outstanding requests to the disk, seek and rotational overheads still exist.
Second, range writes can dramatically improve performance relative to SPTF scheduling, reducing both rotational and seek costs. To achieve the best performance, a file system (or other client) should give reasonably large ranges to the disk: track-sized ranges remove rotational costs, while larger ranges help to noticeably reduce seek time. Although range writes are of greatest utility when there are only a few outstanding writes to the disk, range writes are still useful when there are many.
Third, the presence of reads in a workload obviously reduces the overall effect of range writes. However, range writes can have a noticeable impact even in relatively balanced settings.
Finally, both the EC and HR schedulers perform well, and thus are possible candidates for use within a disk that supports range writes. If one is willing to rewrite the scheduler, HR is the best candidate. However, if one wishes to use the simpler EC approach, one must do so carefully: the full expansion of ranges exacts a high computational overhead.
0 Generally good for positioning time reduction. Could take into account many other features as well, such as track-aligned extents [27].