Table 5.1 summarizes this information for all benchmarks. The column labeled ``Target Procs." is the number of target processors the target program was executed on. Each benchmark was executed for 3 target processor configurations. For example, LU was executed on 4, 8 and 16 processors. The column labeled ``Priv." is the degree of privatization of the simulator, i.e. the maximum number of threads that a host process of the simulation for the corresponding benchmark can support. The three columns labeled ``Targ." contain the size of the executable of the target program and the executable of the simulator for each target processor configuration. In parentheses is the number of host processors on which the simulator was executed. Consequently, the LU benchmark for 4 processors (``Targ. 1") occupies 14M (program + permanent data) per (target) process. After privatization to contain upto 4 threads per process, and linking with MPISIM, it occupies 57M per (host) process. The simulator was executed sequentially, and on 2 and 4 processors. This means that in the sequential simulator, 4 threads were started on the only host process. In the two processor simulator, two threads were started on each host process. In the four processor simulator, only one thread was started per host process. Similarly, the LU benchmark for 8 processors occupies 8M per target process, and after privatization and linking with MPISIM, occupies 32M per host process. Since the LU benchmark was privatized for only 4 threads, the simulator cannot be executed sequentially, since that would require all 8 threads to be started in the same host process.
