P. S. Laursen, "Simple approaches to parallel Branch and Bound", Parallel Computing 19 (1993), p. 143 - 152.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and bound algorithms in parallel processing. In 1984, T.H.Lai and S.Sahni [5] showed the anomalous behavior of parallel branch and bound algorithms. Recently, after putting parallel processing to practical use, results from several implementations on branch and bound algorithms have been presented [2, 6, 7, 9, 10, 12, 14, 15, 16]. The parallelized algorithms (generally called parallel branch and bound algorithms) can be designed in such a way, that there is no interaction between the subproblem currently being solved in parallel and the rest of the existing subproblems (subproblems are generated during the enumeration ....

P.S.Laursen, "Simple approaches to parallel Branch and Bound," Parallel Computing,19 (1993) 143-152.

....and 10 machines posed by Muth and Thompson in 1963 [15] The lack of success in the methods of solving JSS problems to optimality has also initiated much research in the study of heuristic methods. In contrast to other NP hard combinatorial problems like the Quadratic Assignment Problem (QAP) [13], existing heuristics only come within 2 of the optimal value and require substantial computational effort. The situation is illustrated by a recent paper by Brucker et al. 6] in which a B B method is described, which for several instances outperforms the best known heuristics. Since very ....

P.S. Laursen, Simple Approaches to Parallel Branch and Bound, Parallel Computing, 19 (1993) 143--152.

....(extracted from Nugent 20 as a subproblem with symmetries) in 969 seconds, and the Amour Buffa problem of size 20 (which was hitherto unsolved) in 1189 seconds, all on a Cray 2. Of parallel codes solving large QAPs, only a few have been reported by Roucairol [18] Pardalos [9] and Laursen [13], and only Laursen has been able to solve problems from the Nugent test set of size larger than 15 (the solution of Nugent 17 derived from Nugent 20 in appr. 22 hours using a 16 processor T800 transputer system is reported) 3. The Parallel Implementation. In this section we briefly describe the ....

....have a lower bound larger than or equal to the optimal solution value) it is essential for the performance of the algorithm that no processor is idle for a substantial amount of time. Load balancing is a well studied concept, and both simple and more complicated solutions have been proposed, cf. [13, 8]. In the current work we have focused on simple load balancing strategies partly to ease implementation, partly because the problem of avoiding speed up anomalies often addressed through complicated load balancing is not a key issue in the present project. We have experimented with two different ....

[Article contains additional citation context not shown here]

P. S. Laursen, Simple approaches to parallel Branch and Bound, Parallel Computing 19 (1993), p. 143 - 152.

....can be found in the following subsections, relating each implementation to its corresponding theoretic model. Problem reference Integer linear programming shared memory [6] distributed memory [1] Knapsack distributed memory [9] Quadratic assignment (QAP) shared memory [54] distributed memory [32, 47] Scheduling distributed memory [39] Traveling salesperson (TSP) distributed memory [29, 41, 52, 55] Vertex cover (VCP) distributed memory [29, 38] Also, some industrial strength implementations, bearing some resemblance to commercial packages, begin to emerge: Problem reference Mixed ....

....Even the version without dynamic workload sharing exhibited average processor utilization rates between 80 and 90 percent, indicating that the tree search is quite regular in this case. The quantitative workload sharing used is simple, and bring the average processor utilization to 98 percent [32]. R. Luling and B. Monien: the VCP was also used by these authors as an application in order to investigate DDM implementations. The emphasis of their proposal is a quality workload sharing specialized scheme. They proposed some weight functions to express the workload of a processor. The ....

P. S. Laursen. Simple approaches to parallel Branch and Bound. Paral. Comp., 19:143-- 152, 1993.

....iterations, the maximumpath length to a leaf in H is reduced. So, the synchronous algorithm needs at most m n h iterations. However, if we do take the queue maintenance cost into account, we see the difficulties of global best first search. If F is implemented as a centralized priority queue [13] this implies an additional sequential component such that T par 2 Omega (m log m) Besides being a bottleneck, the queue manager s memory might be the limiting factor for the maximal possible problem size. The situation gets only slightly better by pipelining up to O (log m) priority queue ....

P. S. Laursen. Simple approaches to parallel branch and bound. Parallel Computing, 19:143--152, 1993.

....algorithm. The key issues are discussed in the following: ffl State Space Representation: This is a fundamental decision one has to make prior to designing the distributed implementation. One has essentially to decide between a static or a dynamic representation of the states of the problem, [20]. In the former, an initial partitioning is postulated, the corresponding nodes of the branch and bound tree are, somehow, partitioned among the various processors and each processor works exclusively with its assigned pool. In Figure 2 we depict a hypothetical situation in which the initial ....

P. Laursen, Simple approaches to parallel branch and bound, Parallel Computing, 19 (1993), pp. 143--152.

....with respect to both its ability to detect edges and its computational speed. The sequential search used is similar to branch andbound heuristic search techniques. We are currently implementing a parallel version of SEL on a coarsegrained machine. The algorithm, which is similar to that in [15, 16], preserves the edge detection performance while increasing the execution speed nearly linearly with the number of processing elements. Load balancing is performed using a computation estimate developed in [13] A synchronizing step is included for those edges which cross processor partition ....

P. S. Laursen, "Simple approaches to parallel branch and bound," Parallel Computing, vol. 19, pp. 143-- 152, 1993.

No context found.

P. S. Laursen, "Simple approaches to parallel Branch and Bound", Parallel Computing 19 (1993), p. 143 - 152.

No context found.

P.S. Laursen. Simple approaches to parallel branch and bound, Parallel Computing, 19, 1993 143--152.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC