Parallel Computing I (1984) 287-294 287
North-Holland
Performance evaluation of vector
implementations of combinatorial
algorithms *
Celso RIBEIRO
Department of Electrical Engineering~ Catholic Unioersity of Rio de Janeiro, Gavea-Caixa Postal 38063, Rio de
Janeiro 22452, Brazil
Received June 1984
Abstract. We study the performance and the use of vector computers for the solution of combinatorial
optimization problems, particularly dynamic programming and shortest path problems. A general model for
performance evaluation and vector implementations for the problems described above are studied. These
implementations were done on a CRAY-1 vector computer and the computational re:s. ults obtained show (i) the
adequacy of the performance evaluation model and (ii) very important gains concerning computing times,
showing that vector computers will be of great importance in the field of combinatorial optimization.
Keywords. CRAY-1, combinatorial algorithms, vectorization, performance analysis, dynamic programming,
shortest path problems.
1. Introduction
Several recent papers (e.g. Bossavit [3], Kuch et al. [13], Rodrigue [14], Dixon and Patel [8])
are oriented toward the development and the analysis of parallel algorithms for problems
arising in the field of applied mathematics. In most of the eases, computational results have
been obtained by vector implementations on the CRAY-1 computer, one of the fastest
computers currently available.
However, very few results are available concerning combinatorial optimization algorithms,
presently a very important field of research. It was already shown that the complexity of the
resolution of some problems can be reduced by the use of parallel algorithms running on
parallel computers. Among these problems we have the inner product, the product of two
matrices and the sort of the elements of a vector. Cook [5] proved that the class of problems that
can be solved in parallel polynomial time (polynomial time with a polynomial number of
processors) is the same class of problems that can be solved in sequential polynomial time. This
result is very important, since it shows that it is unlikely that we could ever solve hard
NP-complete problems (see Garey and Johnson [10]) in (parallel) polynomial time, unless we
use a nonpolynomial number of processors running in parallel.
Anyway, even if we can not reduce their theoretical complexity, it would be very important
* This research was supported by CNPq (Brazilian Council for Scientific and Technological Development) and the
computer time spent with the use of the CRAY-1 computer was paid by the Department of Applied Mathematics of
CNET (French National Center for Telecommunication Studies). The author acknowledges both institutions for their
financial and administrative help.
016%8191/84/$3.00 © 1984, Elsevier Science Publishers B.V. (North-Holland)