Integrating range writes into file system allocation has proven promising. As desired, range writes can improve performance during file creation while following the constraints of the higher-level file system policies. As much of write activity is to newly created files [4,33], we believe our range-write variant of ext2 will be effective in practice. Further, although limited to data blocks, our approach is useful because traffic is often dominated by data (and not metadata) writes.
Of course, there is much left to explore. For example, partial-file overwrites present an interesting scenario. For best performance, one should issue a range write even for previously allocated data; thus, overwritten data may be allocated to a new location on the disk. Unfortunately, this strategy can potentially destroy the sequentiality of later reads and should be performed with care. We leave this and many other workload scenarios to future work.