Abstract

The widespread use of web applications pushes for faster and more reliable web services. Several techniques have been developed to address web-service reliability, and to provide a better Quality of Service (QoS) for Internet users. A relatively new technique is web-services geographical replication. Geographical replication can be achieved with both server-side and client-side policies. Among these policies, in this paper we contrasted, qualitatively and quantitatively (via simulation), the most promising client-side techniques: one parallel strategy and one single-server strategy. The aim of our study was to determine the pros and cons of the two approaches (single server versus parallel servers) with the aim of identifying the best solutions for content-delivery systems. We analytically quantified the overhead introduced by the selected parallel strategy and proposed two extensions of the parallel strategy to reduce its overhead still maintaining the simplicity of the original approach. The selected strategies are evaluated when: (i) a consistent number of clients adopt the same strategy, (ii) clients download both small and/or big files from the web servers, and (iii) clients are connected to the Internet via a slow-link connection. Results of our analysis pointed out the problems of adopting a parallel strategy in a widespread fashion, and indicated the scenarios in which a parallel strategy may be advantageous. As a result of our study, we proposed a hybrid strategy that combines the parallel and the single server strategies. The hybrid strategy applies either the parallel or the single-server strategy depending on the load and system conditions.

Keywords: Quality of Service; Content delivery policies; Replication; Client-side strategies; Performance evaluation; Simulation

Article Outline

1. Introduction

2. Client-side approaches: the most promising solutions

3. Simulation model

3.1. Servers

3.2. The Internet infrastructure

3.3. Clients

4. Performance comparison

4.1. Impact of the last mile bottleneck

4.2. Parallel strategies refined analysis

4.3. Hybrid strategy

5. Conclusion and further works

Acknowledgements

References

Vitae