Stability of dynamic groups in distributed computing systems

摘要

Large distributed computing systems provide the opportunity for the synergistic use of shared resources; an example model is computational ecosystems. However, as networks continue to expand, it is not obvious that all of the agents within a network should participate in one interacting set. Although smaller groups may forfeit potential synergy, they also avoid much of the complexity and overhead of coordination. The nature of the dynamics is complicated by the inherent delays in networks and the state update schemes applied by the agents. Large intelligent systems must provide for 1) the participation of agents in a smaller group and/or 2) the participation of agents in the systemwide group only a fraction of the time. Both of these behaviors are modeled by a system of differential equations with a discrete distribution of time delays. Stability analysis determines the boundaries for damped and persistent oscillations based on the average of the discrete time delays. In large systems, there is a fixed ratio between these boundaries: eπ2. The results also show that stability improves with diversity in the delays.