The Design And Implementation Of Load Balancing In Heterogeneous Distributed Computing Environment

• Main Authors: Tsai, R.H., 蔡仁豪
• Other Authors: Yang, C.S.
• Format: Others
• Language: zh-TW
• Published: 1997
• Online Access: http://ndltd.ncl.edu.tw/handle/50004012919819025965

碩士 === 國立中山大學 === 資訊工程研究所 === 85 === 　　In heterogeneous distributed computing environment, jobs working at workstations are different and computation power may vary from workstation to workstation. It is possible for some workstations to be overloaded while others are under-loaded or left idle, resulting in poor overall system performance! Thus, the load balancing mechanism is used to determine an appropriate workstation to process a new task for the purpose of fully utilizing processing power of the entire system. 　　We design and implement a load balancing mechanism based on the PVM package. The drawback of PVM's scheduler is that it uses the Rount-Robin method to allocate the coming jobs and may select a heavily node for processing, We modify the scheduler of PVM by using our load balancing algorithm to balance the system load for more efficient system performace. 　　In this thesis, we use the jobs in run queun as the main load index. We use benchmark to normalize the difference of queue length between the workstations with different computing power and define several load levels for each workstation. In information policy, exchange of load information only occurs when a workstation changes its load level to minimize communication costs. According to the load level changes of the wrokstations, a workstation can transfer is task to others at the lower load level to balance the overall system load. 　　Our experimental results have shown that our load balancing algorithm is able to improve performance of PVM environment. Especially at high load, the proposed algorithm yields dramatic performance improvements and performs better than threshold algorithm.