An Enhanced Two-Phases Parallel Loop Self-Scheduling Scheme for PC Clusters and Grid Environments

• Main Authors: Kuan-Wei Cheng, 程冠衛
• Other Authors: Chao-Tung Yang
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/06093212403660645614

碩士 === 東海大學 === 資訊工程與科學系 === 92 === The approaches to deal with scheduling and load balancing on both homogeneous and heterogeneous PC cluster and grid computing system are not mature. Our new method can guarantee when any type of parallel loop scheduling situations happen, they can be properly well-arranged by using our method to achieve a good performance. First of all, the Cluster System Typical Cases are defined for a target system, since many great quantities of performance variances occur because of the different combination of the machines. Therefore we define a frame of relativity by the use of the HINT Performance Analyzer to decide the cluster system for two typical cases comparatively. Following, we must have a proper reaction and an appropriate self-scheduling scheme processed where the system architecture is changeable. The dynamic adjustment of scheduling parameters is used for fitting multiform system architectures to accomplish our system. Now the HINT Performance Analyzer is combined with α self-scheduling scheme, and the dynamic adjustment of scheduling parameters into a whole new approach successfully. The goal of achieving good performance on a parallel loop self-scheduling by using a heuristic approach based upon a two-phase scheme to solve parallel regular loop scheduling problem on both homogeneous and extremely heterogeneous PC cluster and grid computing environment is definitely useful and practicable.