By definition, D-GRAID and other semantically-smart storage systems have more detailed knowledge of the file system that is using them. Embedding a higher degree of functionality within the storage system leads to some concerns on the commercial feasibility of such systems.
The first concern that arises is that placing semantic knowledge within the disk system ties the disk system too intimately to the file system above. For example, if the file system's on-disk structure changes, the storage system may have to change as well. We believe this issue is not likely to be problematic. On-disk formats evolve slowly, for reasons of backwards compatibility. For example, the basic structure of FFS-based file systems has not changed since its introduction in 1984, a period of almost twenty years [28]; the Linux ext2 file system, introduced in roughly 1994, has had the exact same layout for its lifetime. finally, the ext3 journaling file system [45] is backwards compatible with ext2 on-disk layout and the new extensions to the FreeBSD file system [10] are backwards compatible as well. We also have evidence that storage vendors are already willing to maintain and support software specific to a file system; for example, the EMC Symmetrix storage system [11] comes with software that can understand the format of most common file systems.
The second concern is that the storage system needs semantic knowledge for each file system with which it interacts. Fortunately, there are not a large number of file systems that would need to be supported to cover a large fraction of the usage population. If such a semantic storage system is used with a file system that it does not support, the storage system could detect that the file system does not conform to its expectations and turn off its special functionality (e.g., in the case of D-GRAID, revert to normal RAID layout). Such detection can be done by simple techniques such as observing the file system identifier in the partition table.
One final concern that arises is that too much processing will be required within the disk system. We do not believe this to be a major issue, because of the general trend of increasing disk system intelligence [1,37]; as processing power increases, disk systems are likely to contain substantial computational abilities. Indeed, modern storage arrays already exhibit the fruits of Moore's Law; for example, the EMC Symmetrix storage server can be configured with up to 80 processors and 64 GB of RAM [11].