多处理器程序的动态算法.pdfVIP

LOAD BALANCING TECHNIQUES FOR DYNAMIC PROGRAMMING ALGORITHMS ON HYPERCUBE MULTIPTOCESSORS* Steve ASStrate Roger L. Wainwrigbt Department of MathematicalandComputer Sciences The University of Tulsa Abstract ming algorithms is that they trade space for time by storing solutions to subprotdems rather than reconqmting them. In this paper we investigate several load balancing strategies for implemersting the dynamic ~grammirsg technique on a distribut- Dynamic programming is based on the principle of optimality. ed-memory, multiprocessor system. A general method for paral- Simply stated. this means the completion of an optimal se- Ielizing the dynamic programming technique for solving the quence of decisions must be optimaL Hcoce dynamic progra- class of tabular problems is presented. The goal of the package mmingis a powerful technique for solving certain optimization is to relieve a researcher from writing hypercube code from problems. Many problems can be solved using the dynamic scratch for each new algorithm. Currently, simple changes to programming technique. Such problems include context free key routines in the skeleton code is all that is needed. We tested language recognition, fbw shop scheduling, wl knapsack prob- our dynamic programming package on the Matrix Chain produc~