Rapid technological developments have increased computational speed and
power through the exploitation of parallelism.  However, it is often
difficult to test new developments for parallel architectures since 
the architecture may not be available or may be too expensive for the 
researcher to obtain.  A common approach to solving the problem of 
unavailability of the target architecture is to use a simulator to capture 
the behavior of the architectural system under development. In this chapter, 
we present the design of an execution-driven multiprocessor simulator and 
we compare the execution-driven approach to the trace-driven approach.  In 
comparing the two approaches, we observe that the trace-driven approach is 
faster than the execution-driven approach since it does not actually execute 
the threads.  However, the trace-driven approach must store the traces and 
therefore requires more space than the execution-driven approach.  Since the 
execution-driven simulator actually executes the input program, the output 
from the simulated parallel execution can be compared to the output of the 
original sequential program; this comparison is a functional verification of 
the threads produced by our parallelizing compiler.