Service differentiation and load balancing in grid architecture

Abstract

The first part of this paper proposes an architecture to enable service differentiation of jobs on computational grids built using the Globus Toolkit. This feature is absent in the toolkit, which is also unable to manage computation capacity planning among the distributed systems in the grid. The proposed architecture facilitates a front-end web service to accept service-differentiated jobs, allowing the grid to act as a backend Application Service Provider. The second part of this paper proposes an algorithm to load balance the incoming jobs on the various autonomous systems within the grid, taking into consideration the requested QoS level of each job, amount of computation required on the grid, as well as both the existing load and its history to ensure fair and efficient use of computational resources. We conducted extensive experiments that demonstrate the correctness of the algorithm and its limitations. © 2004 IEEE.