As an example of the experimental data obtained, Table 7 shows the context switch times measured in Shark for different hierarchical scheduling strategies. These experiments were executed on a (rather slow) Pentium Celeron running at 300 Mcchz. The test application consisted of 3 servers, each one with a different local scheduler, one with EDF, the second one with Fixed Priority (FP) and the third one with Round Robin (RR). We measured the maximum context switch time between tasks belonging to the same server and between tasks belonging to different servers. The overheads for the hierarchical scheduling structure in Shark, and consequently also for the simpler case in which no hierarchical scheduling is supported, are comparable to the overhead for a normal context switch.