Load balancing strategy and priority of tasks in distributed environments

After showing that lower bound is not the only important parameter to balance subproblems between queues of best-first B&B distributed algorithms, we introduce another notion of priority between subproblems that takes into account not only their lower bound but also their capacity to generate other subproblems. We developed three load balancing strategies using this new notion of priority, with a B&B solution of the Vertex Cover Problem (VCP). The implementation of the VCP was carried out in a network of heterogeneous Unix workstations used as a single parallel computer through a Parallel Virtual Machine (PVM) software system. To show the advantages of the new notion of priority, we solved the VCP for graphs of up to 130 nodes with average degree of 50%. Also, we were able to achieve a remarkable load distribution and a good restriction of communication costs over all the different machines in the system