TY - GEN
T1 - An Extension of Job-Worker Assignment Algorithm for Dynamic Job Migration for User-PC Computing System
AU - Kamoyedji, Ariel Elie Asserehou
AU - Funabiki, Nobuo
AU - Htet, Hein
AU - Zhou, Xudong
AU - Kuribayashi, Minoru
AU - Sugawara, Shinji
N1 - Publisher Copyright:
© 2022 ACM.
PY - 2022/7/29
Y1 - 2022/7/29
N2 - The User-PC computing system (UPC) is a very low cost master-worker model-based distributed computing platform that aims at leveraging idling personal computers (PCs) resources of a group members. In order to do so, the UPC master receives jobs from users and assigns them to available worker PCs, where they are executed using Docker. We have previously devised and implemented an efficient job-worker assignment algorithm considering CPU core utilization, for the UPC system. The latter finds an optimal assignment that minimizes the makespan in the UPC system. In this paper, we extend this algorithm to fully utilize all workers processing power and further reduce the makespan. The proposed method carefully preempts and migrates jobs from their currently assigned worker to another one, based on specific criteria. For evaluation, we conducted experiments using six worker PCs and up to 72 jobs. The extended algorithm could reduce the makespan by up to 65% compared to other existing job scheduling algorithms.
AB - The User-PC computing system (UPC) is a very low cost master-worker model-based distributed computing platform that aims at leveraging idling personal computers (PCs) resources of a group members. In order to do so, the UPC master receives jobs from users and assigns them to available worker PCs, where they are executed using Docker. We have previously devised and implemented an efficient job-worker assignment algorithm considering CPU core utilization, for the UPC system. The latter finds an optimal assignment that minimizes the makespan in the UPC system. In this paper, we extend this algorithm to fully utilize all workers processing power and further reduce the makespan. The proposed method carefully preempts and migrates jobs from their currently assigned worker to another one, based on specific criteria. For evaluation, we conducted experiments using six worker PCs and up to 72 jobs. The extended algorithm could reduce the makespan by up to 65% compared to other existing job scheduling algorithms.
KW - distributed computing
KW - job migration
KW - job scheduling
KW - makespan
KW - optimization
KW - UPC
UR - http://www.scopus.com/inward/record.url?scp=85140794310&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85140794310&partnerID=8YFLogxK
U2 - 10.1145/3556223.3556250
DO - 10.1145/3556223.3556250
M3 - Conference contribution
AN - SCOPUS:85140794310
T3 - ACM International Conference Proceeding Series
SP - 175
EP - 183
BT - Proceedings of the 10th International Conference on Computer and Communications Management, ICCCM 2022
PB - Association for Computing Machinery
T2 - 10th International Conference on Computer and Communications Management, ICCCM 2022
Y2 - 29 July 2022 through 31 July 2022
ER -