Recent research results and infrastructure efforts demonstrate the potential effectiveness of large-scale distributed computing. Effective scheduling based on empirically verifiable models has emerged as a key factor in these successes. Moving to a new truly global computing capability will similarly depend critically on new models and scheduling techniques. The development and instantiation of models for computing platforms are critical first steps for studying the feasibility and scalability of applications, as well as for developing methodologies to enhance application performance. Our focus, in this proposal, is on the modeling technologies that will enable performance engineering of such global computations. Specifically, these new modeling technologies will support resource characterization and scheduling techniques thereby enabling robust, predictable performance.