Although the percentage overhead for some of the microbenchmark results is large, the real difference in absolute times is typically quite small and becomes insignificant for macro operations, as shown by the results in Section 6.2. Furthermore, these results must be viewed as an upper bound on the performance overhead, since neither the AVC nor the security server implementation have been optimized. Other known areas where the performance could be improved include making better use of AVC entry references and improving the AVC locking scheme.