The implementation of this function was run on 1, 2 and 4 processors, the times for the computation of f(8), a tree of depth at most 8, containing 1701 nodes. Fig. 7.2 shows the time taken for the evaluation to run, the error bars representing the range in time taken for the various processors involved with the computation. Naturally, the run time of a set of parallel processes is taken to be the time for the longest process to complete, given that they all begin at the same time. The two processor version therefore runs in 60% of the time for the one processor computation, a significant speedup. When we run using four processors, however, the total time increases drastically. If we apply the analysis developed earlier to this particular example, we have k=4 and r=2, so we can expect no more that two-way parallelism. So, as predicted, we can achieve no more than two way parallelism with this particular example.