A new parallel algorithm for the knapsack problem and its implementation on a hypercube

A new parallel algorithm is presented for the 0/1 knapsack problem. On a hypercube of n processors, the algorithm runs in time O(mc(log n)/n), where m is the number of objects and c is the knapsack size. The best previous known hypercube algorithm takes time O(mc/n+c log n+c/sup 2/). The new algorithm has been implemented on the Connection Machine and experimental results show that it performs very well for a wide range of problem sizes.<<ETX>>