Formal analysis of an agent-based optimisation strategy for Data Grids

Abstract

In a world-wide computational Grid, users typically want their jobs to be executed as fast as possible, while the goal of a Grid infrastructure is to assure specific quality of service for all users. In order to reconcile these apparently contrasting goals, the authors proposed an economy-based strategy to be used in a Data Grid for efficient access to and distribution of data replicas needed by Grid jobs [20]. The strategy is based on an economic model where data-seeking agents negotiate optimal prices for exchanging data files with data-storing agents. Data-storing agents also try to keep on their Grid sites the most useful files. In previous works, performance of the economic model has been empirically studied via simulations conducted using the Data Grid simulator OptorSim [11,18]. In this paper the problem of efficient access to and distribution of replicas is introduced and the economic model is described. A formalisation of the auction protocol basis of the economic model is then presented and formal proof of some properties (namely, that it is deadlock-free and always terminates) is provided. The auction protocol is modelled using Petri nets, a formal and graphical language that is well suited for modelling concurrent distributed systems.