Online Cache Modeling For Commodity Multicore Processors

Modern chip-level multiprocessors (CMPs) contain multiple processor cores sharing a common last-level cache, memory interconnects, and other hardware resources. Workloads running on separate cores compete for these resources, often resulting in highly-variable performance. To improve fairness and performance, it is helpful to co-schedule workloads having minimal cache and other forms of resource contention. In this work, we develop several cache modeling techniques to help make informed resource management decisions.Using only commonly-available performance counters on existing processors, we introduce an efficient online technique for estimating the cache occupancies of software threads. We derive an analytical model that considers the impact of set-associativity, line replacement policy, and memory locality effects. We demonstrate the effectiveness of occupancy estimation with a series of CMP simulations using SPEC benchmarks. Our occupancy estimation technique is currently being used to develop online utility functions, such as miss-ratio curves (MRCs), which capture performance impacts as a function of resource usage. We are leveraging both online cache occupancy estimation and MRC construction in our ongoing studies of cache-aware scheduling.