Load Balancing and Workstation Autonomy on Amoeba

This paper presents the results of a load balancing study carried out on the distributed operating system Amoeba. The study is intended to investigate the e ectiveness of di erent load balancing methodologies on a workstation-based system, specifying job initiation vs. process migration. The results indicate that both methods can improve system performance, such as response time. The results also show that job initiation plays a more important role in a load balancing scheme than process migration for the process migration mechanism used in this work. A number of load balancing algorithms, both centralized and distributed, have been studied in this work. The results of our experiments show that a centralized algorithm outperforms a distributed one in performance and scalability. We also discuss the trading technique used in this study, which allows the owner of a workstation to decide whether join or leave load balancing dynamically. We conclude with a summary of our experiences and suggestions for