We take a two-pronged approach to answering the questions posed in Section 1. First we examine node-level measurement data about individual resources, independent of any application model. We then compose this data about available node resources with simple models of application resource demand derived from popular PlanetLab services. This second step allows us to evaluate how space-and-time-varying sliver resource demand interacts with space-and-time-varying free node resources to determine the potential benefits of informed sliver placement and migration, as well as the impact of optimizations to a migration service.
We ask our questions from the perspective of a single application at a time; that is, we assume all other applications behave as they did in the trace. Thus, our approach combines measurement, modeling, and simulation, but the simulation does not consider how other users might react to one user's placement and migration decisions. We leave to future work developing a multi-user model that might allow us to answer the question ``if multiple PlanetLab users make placement and migration decisions according to a particular policy, how will the system evolve over time?''
