This paper develops a framework to model the performance of parallel applications executing in a shared network computing environment. For sharing of a single computation node or network link, the actual performance is predicted, while for sharing of multiple nodes and links, performance bounds are developed. The methodology for building such a shared execution performance model is based on monitoring an application’s execution behavior and resource usage under controlled dedicated execution. The procedure does not require access to the source code and hence can be applied across programming languages and models. We validate our approach with experimental results with NAS benchmarks executed in different resource sharing scenarios on a small cluster. Applicability to more general scenarios, such as large clusters, memory and I/O bound programs and wide are networks, remain open questions that are included in the discussion. This paper makes the case that understanding and modeling application behavior is important for resource allocation and offers a promising approach to put that in practice.