PACORA: Performance Aware Convex Optimization for Resource Allocation

Resource allocation is the dynamic allocation and de-allocation of processor cores, memory pages, and various categories of bandwidth to client subcomputations (e.g processes within an operating system) that compete for those resources. Historically, resource allocation has been rather unsystematic, and now the assumptions underlying the traditional strategies no longer hold. The existence of parallel software and multiple heterogeneous processor cores, the requirement to maintain quality of service agreements and responsiveness, and the need to limit power and energy consumption (especially in mobile systems) are inadequately addressed by the status quo. A new approach to resource allocation is described that addresses all of these needs and that can be framed as a convex optimization problem; the result is that an optimal solution will exist and be unique with no local extrema. As a result, rational, fast, and fully incremental solutions to the resource allocation problem become feasible.