The International Conference for High Performance Computing, Networking, Storage and Analysis
Semantically Ordered, Parallel Execution of Multiprocessor Programs.
Student: Gagan Gupta (University of Wisconsin-Madison)
Advisor: Gurindar S. Sohi (University of Wisconsin-Madison)
Abstract: Conventional wisdom says that we must abandon the logical order between computations to obtain high performance from multiprocessor programs.
Unfortunately, the resulting nondeterminism complicates parallel programming, resource management, recovery from exceptions, and parallel system design in general. Conversely, uniprocessor systems have long benefited from ordered programs. Order leads to deterministic execution, which does not suffer from the shortcomings of nondeterminism. This work shows that an ordered approach can also be applied to multiprocessor programs, simplifying parallel system design, but without compromising performance.