Ruurd Wiegers - Academia.edu (original) (raw)
Uploads
Papers by Ruurd Wiegers
Proceedings of the Seventh International Conference on Parallel and Distributed Computing Systems (PDCS'94), 1994
Parallel programming problems which solve multiple branch-and- bound nodes in parallel, often sh... more Parallel programming problems which solve multiple branch-and- bound nodes in parallel, often show a speedup which is not monotone increasing, and sometimes even show a slowdown.
In this paper we compare the performance of two versions of a parallel synchronous branch-and-bound solver for 0-1 integer programming problems. The first version runs on a transputer system using the VCR (Virtual Channel Router), the second version on a network of UNIX-workstations using the PVM (Parallel Virtual Machine) software.
The parallel branch-and-bound solver described in this paper uses the parallelism in the branch-and-bound tree, i.e. multiple branch-and-bound nodes are solved in parallel as linear programming problems using a farmer-worker parallelization of the branch-and-bound algorithm. Depth first scheduling of the branch-and-bound nodes is used. The obtained speedup is theoretically monotone increasing.
Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-b... more Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-bound for mixed integer programming if we use the parallelism in the branch-and-bound tree. The machine model assumed is a distributed memory parallel computer. The program model is farmer-worker, in which the farmer creates the branch-and-bound tree, while the workers calculate the linear programming solution of the branch-and-bound nodes assigned to them by the farmer. We will see that for a depth first search we are able to guarantee the speedup to be larger than or equal to one, while superlinear speedup can still occur. The described measures to improve the speedup of the parallel branch-and-bound solver for mixed integer programming have been tested using problems from MIPlib. Test results show that by using the described approach the theoretical and real speedups are larger than one and monotone increasing, and sometimes even superlinear.
Massively Parallel Processing Applications and Development, 1994
Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-b... more Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-bound for mixed integer programming if we use the parallelism in the branch-and-bound tree. The machine model assumed is a distributed memory parallel computer. The program model is farmer-worker, in which the farmer creates the branch-and-bound tree, while the workers calculate the linear programming solution of the branch-and-bound nodes assigned to them by the farmer. We will see that for a depth first search we are able to guarantee the speedup to be larger than or equal to one, while superlinear speedup can still occur. The described measures to improve the speedup of the parallel branch-and-bound solver for mixed integer programming have been tested using problems from MIPlib. Test results show that by using the described approach the theoretical and real speedups are larger than one and monotone increasing, and sometimes even superlinear.
Ifip Wg10 3 Publications, 1993
Proceedings of the Seventh International Conference on Parallel and Distributed Computing Systems (PDCS'94), 1994
Parallel programming problems which solve multiple branch-and- bound nodes in parallel, often sh... more Parallel programming problems which solve multiple branch-and- bound nodes in parallel, often show a speedup which is not monotone increasing, and sometimes even show a slowdown.
In this paper we compare the performance of two versions of a parallel synchronous branch-and-bound solver for 0-1 integer programming problems. The first version runs on a transputer system using the VCR (Virtual Channel Router), the second version on a network of UNIX-workstations using the PVM (Parallel Virtual Machine) software.
The parallel branch-and-bound solver described in this paper uses the parallelism in the branch-and-bound tree, i.e. multiple branch-and-bound nodes are solved in parallel as linear programming problems using a farmer-worker parallelization of the branch-and-bound algorithm. Depth first scheduling of the branch-and-bound nodes is used. The obtained speedup is theoretically monotone increasing.
Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-b... more Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-bound for mixed integer programming if we use the parallelism in the branch-and-bound tree. The machine model assumed is a distributed memory parallel computer. The program model is farmer-worker, in which the farmer creates the branch-and-bound tree, while the workers calculate the linear programming solution of the branch-and-bound nodes assigned to them by the farmer. We will see that for a depth first search we are able to guarantee the speedup to be larger than or equal to one, while superlinear speedup can still occur. The described measures to improve the speedup of the parallel branch-and-bound solver for mixed integer programming have been tested using problems from MIPlib. Test results show that by using the described approach the theoretical and real speedups are larger than one and monotone increasing, and sometimes even superlinear.
Massively Parallel Processing Applications and Development, 1994
Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-b... more Abstract This paper describes the measures needed to improve the speedup of parallel branch-and-bound for mixed integer programming if we use the parallelism in the branch-and-bound tree. The machine model assumed is a distributed memory parallel computer. The program model is farmer-worker, in which the farmer creates the branch-and-bound tree, while the workers calculate the linear programming solution of the branch-and-bound nodes assigned to them by the farmer. We will see that for a depth first search we are able to guarantee the speedup to be larger than or equal to one, while superlinear speedup can still occur. The described measures to improve the speedup of the parallel branch-and-bound solver for mixed integer programming have been tested using problems from MIPlib. Test results show that by using the described approach the theoretical and real speedups are larger than one and monotone increasing, and sometimes even superlinear.
Ifip Wg10 3 Publications, 1993