next up previous contents index
Next: 13.6 Coherent Parallel C Up: 13.5 ASPAR Previous: 13.5.6 Dynamic Data Distribution

13.5.7 Conclusions

The success of ASPAR relies on two crucial pieces of technology:

It is interesting that neither of these is the result of any extensions to existing compiler technology but are derived from our experience with parallel computers. This is consistent with our underlying philosophy of having ASPAR duplicate the methods which real programmers use to successfully parallelize code by hand. Obviously not all problems are amenable to this type of automatic parallelization but we believe that of the cases discussed in the opening paragraphs of this section we can usefully address all but the ``Extremely Difficult.''

In the simpler cases, we believe that the goal of eliminating the role of ``human error'' in generating correctly functioning parallel code has been accomplished.

The price that has been paid, of course, is the requirement for extremely smart runtime systems. The use of Express as the underlying mechanism for ASPAR has proved its value in addressing the simpler types of decomposition scheme.

The development of the dynamic data-distribution mechanisms required to support the more complex applications has led to a completely new way of writing, debugging,  and optimizing parallel programs which we believe will become the cornerstone of the next generation of Express systems and may revolutionize the ways in which people think about parallel processing.

Guy Robinson
Wed Mar 1 10:19:35 EST 1995