A Performance Comparison of two Decomposition Techniques for Multilevel Secure Database Systems

In this paper, we evaluate and compare the performance of two decomposition techniques for multilevel secure database systems: the original decomposition proposed in SeaView and the ``Novel'''' decomposition proposed by Jajodia and Sandhu. While SeaView decomposes a relation into partitions using horizontal and vertical fragmentation, on the basis of tuple and attribute security class, Novel uses only horizontal fragmentation to break multilevel relations into single-level ones; as a result, Novel claims to lower the cost of view reconstruction. As part of this study, we derive expressions for both the storage and I/O costs for view reconstruction, select and project operations for both techniques. We conclude that when only a small percentage of attributes of low-level tuples are updated by a high-level user, the storage overhead in Novel is significantly high compared to SeaView; it is comparable in other cases. However, unlike SeaView, Novel does not generate extraneous tuples during view reconstruction. This not only results in reduced I/O cost (especially when the percentage of updates is higher) but also eliminated the additional information-retrieval related problems posed by SeaView.