Social grid agents

Abstract

The problem of resource allocation in Grid computing has been actively tackled by the scientific community for some years; its complexity is in meeting the expectations of different actors with different concepts of optimality within an environment divided into geographically dispersed and different administrative domains and thus unsuited to supporting centralized management systems. The complexity of the problem is further increased by the fact that the scientific community produced different Grids with different standards, focuses and technologies. This problem closely resembles that of an economy and has consequently been investigated with economic paradigms, leading to models that are often tailored to one or more specific situation. This thesis speculates that an approach reflecting the social structures that support economic models could allow different models to co-exist as they do in the real world; this thesis also speculates that a successful solution should be able to encompass different Grids both present and future. In order to achieve this, the modelling of the different actors and their behaviour is inspired by some behaviours that are common in societies; to encompass different grids, the environment of the actors is modelled as a metagrid: a conceptual space divided into three regions: the different existing middlewares, the metagrid region and border regions at the intersection of the two where interoperability issues are tackled by a combination of abstraction and translation. This thesis proposes that resource allocation be modelled as the intersection of different topologies where actors interact with each other to enforce different allocation philosophies. These actors are modelled with agents termed Social Grid Agents. 2 The proposed model is based on four different kinds of topologies: a Production Topology where services are performed, a Social Topology where agents that control the production process engage in social and economic processes, a Control Topology where the agents belonging to the Social Topology control those belonging to the Production Topology, and a Value and Price Topology that connects social and production agents to determine the value and price of a resource. Social Grid Agents communicate with messages and their behaviour is defined by sets of policies that determine how they relate to each other. The need for an efficient and expressible native language with which the agents express policies and messages led to the decision to use a functional language as a base for the agent’s native language. An architecture is defined at an abstract and concrete level and a prototype that can encompass different social relationships has been implemented and tested in a series of experiments to evaluate the efficiency, scalability and behaviour of the Social Grid Agents.