Task Scheduling And Load Balancing Methods In NOW

NOW(Network Of?Workstations) is a network parallel computing environment consists of multi-computers connected by high speed networks and special network operating system, which can be used to large scale parallel processing and load balancing"(This definition is made by the NOW project group of University of California, Berkeley). With rapid improvement of computing capability of the low-end computers (e.g. PCs) and with continuous progress in computer networks and communication techniques, NOW has been playing more and more important role in parallel computing. First of all, cost-effectiveness of NOW makes it a favorite choice for parallel computing compared with expensive mainframes and i.e super-computers. Secondly, a fringe benefit of utilizing NOW systems is to keep a system flexible, not only as a close-coupled system but also as an individual system to provide service for different kinds of user and processing job. Task scheduling and load balancing are very important issues for NOW designers, which will have direct impact on processing efficiency and resource utilization. This is exactly the reason why the author has devoted tremendous efforts on these topics in his PhD research work. First of all, a comprehensive review of existing works and potential development trend in task scheduling and load balancing field, especially in NOW, is given, merits and shortcomings concerning each method are analyzed in some detail. Based on the existing works, this dissertation deeply explores the characteristics and behavior of task scheduling and load balancing in NOW, proposes and creates ten creative task scheduling and load balancing methods in NOW such as the sender and receiver combinedly initializing(SRI), process migration among nodes, runtime incremental concentrated scheduling and so on. These new ten methods are proved to be high practical. For each new method proposed by the author, the dissertation introduces the idea detailedly, describes the implementing way clearly, and gives the experimental results. These ten creative methods are: I. The task scheduling and load balancing method of the sender and receiver combinedly initializing(SRI); 2. The method of process migrating among nodes; 3. The sender initializing strategy supporting process migration among nodes; 4. The receiver initializing strategy supporting process migration among nodes; 5. The coscheduling method not supporting process migration among nodes; 6. The coscheduling method supporting process migration among nodes; 7. The implicit coscheduling method supporting process migration among nodes; 8. Runtime incremental concentrated scheduling in NOW; 9. Runtime incremental concentrated scheduling based on tree-liking cluster structure; 10. Intelligentized task scheduling and load balancing methods based on MAS (Multi-Agents System). At the end of this dissertation, the author applies the task scheduling and load balancing methods in NOW such as the sender initializing strategy, the receiver initializing strategy and the above sender and receiver combinedly initializing strategy, runtime incremental concentrated scheduling method, intelligentized task scheduling and load balancing method etc. to a parallel Radar Radiating-Source...