Simulating a computational grid with networked animat agents

Computational grids are now widespread, but their large-scale behaviour is still poorly understood. We report on some calculations of loading, scaling and utilisation behaviours of computational grids, based on simulations. We employ animat agents on a topologically detailed graph representing a grid overlay network. Agents are used to represent computational jobs, users and resources. We can obtain realistic behaviours by endowing user agents with time-varying microscopic behaviour patterns. We are able to study the static flow and dynamical macroscopic properties of the network including emergent pathological behaviours and other anomalies that arise when parts of the network become temporarily unavailable. Our model is based on graph theory with various attributes decorating the edges and nodes which have physical locations. We develop some overall grid utility metrics that can be analysed. Our work suggests that grids do need to be treated as complex adaptive systems.