Home // ICSEA 2012, The Seventh International Conference on Software Engineering Advances // View article

Explicit use of Working-set Correlation for Load-balancing in Clustered Web Servers

Authors:
Stoyan Garbatov
João Cachopo

Keywords: clustered web servers; load balance; locality awareness; Latent Dirichlet Allocation; scalability; performance

Abstract:
This work presents a new load balancing policy for clustered web server systems. With this policy, every node in the cluster is responsible for dealing with a particular subset of request types. All requests are partitioned into disjoint groups, according to the domain data contained in their working sets. The composition of the working sets is established by using automatic data access pattern analysis and prediction techniques. Latent Dirichlet Allocation is employed as the partitioning algorithm to maximize the similarity (correlation) between the working sets of the requests placed in the same group. The working-set correlation is used explicitly with the aim of improving the data locality of all functionality that is to be executed on a given node, leading to more efficient use of computational resources and, ultimately, to increased performance. The work was validated on the TPC-W benchmark.

Pages: 699 to 704

Copyright: Copyright (c) IARIA, 2012

Publication date: November 18, 2012

Published in: conference

ISSN: 2308-4235

ISBN: 978-1-61208-230-1

Location: Lisbon, Portugal

Dates: from November 18, 2012 to November 23, 2012