| Summary: | 碩士 === 國立高雄第一科技大學 === 系統資訊與控制研究所 === 101 === During the development stage, a developed software system is focused on automation and correctness of its outputs. However, output efficiency of the system gets worse while increasing of data items and amounts in the operating stage. In particular, a time-out issue occurs when the limited resources of the system are competed for processing increasing data by various jobs which are owned by different roles for different time constraints.
This work proposes the particle swarm optimization (PSO) based parallel processing architecture to schedule and execute the system jobs for satisfying the multi-objective: roles, works, and due dates. In application, we enhanced the materiel control system to present efficiency of the architecture. For a general material control system, if the vendors and product groups of material items are independent, the tasks of material reports can be separately executed according to vendors or product groups. Hence, an execution plan in the available virtual machines can be scheduled by the PSO according to the needed execution times of the tasks. The task data are preloaded from the database to the hash tables of the corresponding virtual machine by following the execution plan to form the data cache. Using the distributed computing method Map Reduce, the tasks are mapped to the available virtual machines for parallel processing. Finally, the results of the tasks are integrated and reduced for generating the final reports.
In execution efficiency, the report jobs of a fabless material management case, which handles 21 fabs for outsourcing manufacturing 798 product groups, indicate that each original jobs spent 35 min. on average while executing on a computing node. Using the proposed processing architecture, the mean execution time of report jobs is 2.5 min. which is decreased 93% of the original time while executing on four virtual machines. Especially, the constraint of the job execution time (3 min.) by increasing the numbers of virtual machines while material data are increased. Hence, the proposed architecture is proven with data extendibility and time-constraint insurance.
|
|---|
