系統的シャフルを用いた並列型分枝限定法

マルチプロセツサタスクスケジューリンダへの応用

抄録

The most general technique that can be used to solve a wide variety of combinatorial extremumsearch problems, such as the task scheduling for paraller processing, is the branch-and-bound algorithm. However, since these problems are nondeterministic polynomial (NP)-complete, the solution derived within an appropriate searching time is not always optimal.
In this paper, from the standpoint of avoiding local minimum solutions. we propose a practical scheduler which searches a wide area of the search tree using plural searching strategies based on the latest starting time and performing a systematic shuffling of the proority list for node selections.
Our computer simulation results demonstrate that the systematic shuffling yields to dramatic acceleration in obtaining the candidate solutions. An application of the proposed scheduler to the simulation of the impact drop of a roll milling machine indicates that the total idle time of the processors can be decreased by a factor of ninety when compared to the conventional scheduling approachs. The observation of a mechanical manipulator control application reveales that the minimum number of processors for reaching the critical path length is five, with the conventional scheduling requiring seven.