TY - JOUR
T1 - Independent and cooperative parallel search methods for the generalized assignment problem
AU - Asahiro, Yuichi
AU - Ishibashi, Masahiro
AU - Yamashita, Masafumi
N1 - Funding Information:
This work was supported in part by Grant-in-Aid for Scientific Research (KAKENTHI), No.12878052, No.14380145, and No.14085204 from The Ministry of Education, Culture, Sports, Science and Technology of Japan and Japan Society for the Promotion of Science.
PY - 2003/4
Y1 - 2003/4
N2 - The generalized assignment problem is a representative NP-hard problem, for which many heuristic algorithms are known. In this article, two parallel heuristic algorithms are proposed, which are based on the ejection chain local search (EC) proposed by Yagiura et al. One is a simple parallelization called multistart parallel EC (MPEC) and the other is cooperative parallel EC (CPEC). In MPEC each search process independently explores search space while in CPEC search processes share partial information to cooperate with each other. The experimental results with 9 computers for large benchmark instances show that (1) MPEC and CPEC, respectively, run twice and 4 times faster than EC, and (2) compared to EC, the difference in quality between obtained solutions and theoretical lower bounds is reduced to 3/4 and 2/3 by MPEC and CPEC, respectively. It is said that these methods give us full benefit of parallelization, speedup and improvement for quality of solutions.
AB - The generalized assignment problem is a representative NP-hard problem, for which many heuristic algorithms are known. In this article, two parallel heuristic algorithms are proposed, which are based on the ejection chain local search (EC) proposed by Yagiura et al. One is a simple parallelization called multistart parallel EC (MPEC) and the other is cooperative parallel EC (CPEC). In MPEC each search process independently explores search space while in CPEC search processes share partial information to cooperate with each other. The experimental results with 9 computers for large benchmark instances show that (1) MPEC and CPEC, respectively, run twice and 4 times faster than EC, and (2) compared to EC, the difference in quality between obtained solutions and theoretical lower bounds is reduced to 3/4 and 2/3 by MPEC and CPEC, respectively. It is said that these methods give us full benefit of parallelization, speedup and improvement for quality of solutions.
UR - http://www.scopus.com/inward/record.url?scp=0037810924&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0037810924&partnerID=8YFLogxK
U2 - 10.1080/1055678031000107105
DO - 10.1080/1055678031000107105
M3 - Article
AN - SCOPUS:0037810924
SN - 1055-6788
VL - 18
SP - 129
EP - 141
JO - Optimization Methods and Software
JF - Optimization Methods and Software
IS - 2
ER -