Flexible distributed computing with volunteered resources

Abstract

Nowadays, computational grids have evolved to a stage where they can comprise many volunteered resources owned by different individual users and/or institutions, such as desktop grids and volunteered computing grids. This brings benefits for large-scale computing, as more resources are available to exploit. On the other hand, the inherent characteristics of the volunteered resources bring some challenges for efficiently exploiting them. For example, jobs may not be able to be executed by some resources, as the computing resources can be heterogeneous. Furthermore, the resources can be volatile as the resource owners usually have the right to decide when and how to donate the idle Central Processing Unit (CPU) cycles of their computers. Therefore, in order to utilise volunteered resources efficiently, this research investigated solutions from different aspects. Firstly, this research proposes a new computational Grid architecture based on Java and Java application migration technologies to provide fundamental support for coping with these challenges. This proposed architecture supports heterogeneous resources, ensuring local activities are not affected by Grid jobs and enabling resources to carry out live and automatic Java application migration. Secondly, this research work proposes some job-scheduling and migration algorithms based on resource availability prediction and/or artificial intelligence techniques. To examine the proposed algorithms, this work includes a series of experiments in both synthetic and practical scenarios and compares the performance of the proposed algorithms with existing ones across a variety of scenarios. According to the critical assessment, each algorithm has its own distinct advantages and performs well when certain conditions are met. In addition, this research analyses the characteristics of resources in terms of the availability pattern of practical volunteer-based grids. The analysis shows that each environment has its own characteristics and each volunteered resource’s availability tends to possess weak correlations across different days and times-of-day.