Analysis of scheduling policies in a user-level prioritization service

Most desktop applications execute on commodity operating systems and networks that support end-user requirements on a best-effort basis. As a result, the end-user cannot indicate to the underlying system the priority of networked applications when they are executed concurrently. Consequently, the performance of these applications is degraded non-deterministically based on the availability of resources and furthermore the degradation is not in the order and to the extent the end-user would want. This unpredictable degradation frequently affects the usefulness of a user session when multiple networked applications are executed simultaneously. We designed and implemented a prioritization service that allows an end-user to prioritize applications according to the requirements of an individual session. The service has been implemented in user-space and incorporates several scheduling disciplines such as Weighted Fair Queuing, Class Based Queuing, and Priority Queuing. We conducted a performance evaluation of the impact of the scheduling choices on the networked applications. Our empirical performance results demonstrate that end-user applications can be prioritized in the order the end-user chooses depending on the scheduling policies.