A new generation of load sharing algorithms: the semi-adaptive load sharing algorithm

Abstract

Distributed systems have become a popular computing environment. Due to their high potentials in improving performance and resource sharing, the evolution and maturing of technologies such as networks and computer hardware, have led to more realizations of distributed computing capabilities. One of the key factors that impacts the system's performance is how load sharing is handled. Most of the work done in load sharing algorithms focused on either one of two approaches: the static approach, or the dynamic approach. Static load sharing is characterized with its simplicity and minimal overhead induced on the system when performing load distribution. This is attributed to the fact that their load sharing decisions do not require any information exchange between nodes. However, static algorithms are not capable of adapting to traffic changes. Hence, this will affect the distributed system adaptability and stability features against load alteration. The second approach in load sharing algorithums is the dynamic approach. In this approach, nodes base their sharing decisions on the knowledge of the status on other nodes. This knowledge enables nodes to make improved decisions when selecting a remote execution site. Therefore, these algorithms provide the distributed system with adaptability and stability. Nevertheless, information exchange between nodes has proven to be expensive. The problem lies in the communication and processing overheads which these algorithms impose on the system. This research work presents a Semi-Adaptive load sharing approach that attempts to lay a common ground between the dynamic and the static approaches. This approach provides the system with adaptability features at are supplied with tools to control the level of adaptability and information exchange in the system. It is argued that this approach is adaptable, scalable, stable, fault tolerant, and will generate low and controlled system overheads. Two implementations of the Semi-Adaptive approach are presented: The Controller Load Sharing Protocol (CLSP) and The Pool Load Sharing Protocol (PLSP). The study of CLSP and PLSP include simulating their performance under various conditions. The simulation include comparing the performance of these two protocols against two load sharing algorithms representing the static and dynamic families. Furthermore, the properties of CLSP and PLSP are investigated. Finally, based on the study and analysis of both protocols, the research work proposes possible areas of improvement in the final conclusion.