Overall, we were surprised by the many similarities we found in implementing D-GRAID underneath ext2 and VFAT. For example, VFAT also overloads data blocks, using them as either user data blocks or directories; hence Alexander must defer classification of those blocks in a manner similar to the ext2 implementation.
However, there were a few instances where the VFAT implementation of D-GRAID differed in interesting ways from the ext2 version. For example, the fact that all pointers of a file are located in the file allocation table made a number of aspects of D-GRAID much simpler to implement; in VFAT, there are no indirect pointers to worry about. We also ran across the occasional odd behavior in the Linux implementation of VFAT. For example, Linux would write to disk data blocks that were allocated but then freed, avoiding an obvious and common file system optimization. Although this was more indicative of the untuned nature of the Linux implementation, it served as yet another indicator of how semantic disks must be wary of any assumptions they make about file system behavior.