| Summary: | 碩士 === 銘傳大學 === 資訊管理研究所 === 88 === The traveling salesperson problem is a classic and famous problem in combinatorial optimization research. This problem can be adapted to an entire class of NP-complete problems. To find an efficient way to resolve those problems relating to the traveling salesperson problem would be helpful. That could be applied to network application、industrial production、transportation.
The branch-and-bound strategy is a well-known and efficient approach to solve many combinatorial problems. Fruitful results about branch-and-bound algorithms can be found in the literature since the 1960's. Consequently, much efforts have been made on the study of parallel branch-and-bound algorithms after the late 1980's.
In this research, a more generalized strategy, i.e., the “multithread branch-and-bound”, is proposed. More than one thread can be executed in a single processor concurrently in our idea. Each thread might correspond to a part of a branch in the input problem’s branching tree whose lower and upper bounds are computed. A message-passing share-memory is devised for threads to instantly access the mostly updated upper bound to bound unnecessary branches inside each thread. As a way, these threads can cooperatively solve a problem in branch-and-bound manners more efficiently.
We build an extensible architecture was named “cluster of computers”. We develop the algorithm of “multithread branch-and-bound” in this distributed computing environment, which was constructed with personal computer and TCP/IC network protocol. We will make full use of PC computing capacity.
|
|---|
