A Two-Tier Load Balancing Scheme for Distributed Systems

Abstract

Workstations within a distributed system often contain powerful processors, although on average, the processing demands of the user are far less than the capacity of the workstation they are using. Load balancing uses the ability to remotely execute processes to allow for the spreading of the computational load over the workstations within a distributed system. This paper presents a summary of the issues and characteristics that must be considered when developing a load balancing algorithm. It looks at the development of a load balancing mechanism using techniques from a number of important solutions within the area, with emphasis being placed on scalability, stability, fairness of service, efficiency and a move away from an “all or nothing” approach where a single load threshold mark is used, using a High-Low scheme [1]. The paper provides both a theoretical and performance based analysis of the algorithm in order to provide a fair evaluation. Results show that improvements in user response time are evident when idle/lightly loaded processor(s) are available in order to accept remote processes.