Communist, utilitarian, and capitalist cache policies on CMPs

Abstract

As chip multiprocessors (CMPs) become increasingly mainstream, architects have likewise become more interested in how best to share a cache hierarchy among multiple simultaneous threads of execution. The complexity of this problem is exacerbated as the number of simultaneous threads grows from two or four to the tens or hundreds. However, there is no consensus in the architectural community on what "best" means in this context. Some papers in the literature seek to equalize each thread's performance loss due to sharing, while others emphasize maximizing overall system performance. Furthermore, the specific effect of these goals varies depending on the metric used to define "performance".In this paper we label equal performance targets as Communist cache policies and overall performance targets as Utilitarian cache policies. We compare both of these models to the most common current model of a free-for-all cache (a Capitalist policy). We consider various performance metrics, including miss rates, bandwidth usage, and IPC, including both absolute and relative values of each metric. Using analytical models and behavioral cache simulation, we find that the optimal partitioning of a shared cache can vary greatly as different but reasonable definitions of optimality are applied. We also find that, although Communist and Utilitarian targets are generally compatible, each policy has workloads for which it provides poor overall performance or poor fairness, respectively. Finally, we find that simple policies like LRU replacement and static uniform partitioning are not sufficient to provide near-optimal performance under any reasonable definition, indicating that some thread-aware cache resource allocation mechanism is required.

A Flexible Framework for Throttling-Enabled Multicore Management (TEMM)

“…We are aware that there are several possible definitions of fair use of shared resources [16]. The particular choice of fairness measure should not affect the main purpose of our work.…”

Cooperative Caching for Chip Multiprocessors

“…Rafique et al [125] and Petoumenos et al [121] proposed spatially fine-grained partitioning support, which can be used by various partitioning policies (such as miss rate reduction, fair caching and QoS provision). Hsu et al [66] studied various partitioning metrics and policies. Their study focused on three caching paradigms (communist caching for fairness, utilitarian caching for overall throughput and uncontrolled capitalist caching), and recognized the difficulties to improve both overall throughput and fairness using a single partitioning scheme.…”

“…Measurement information can be gathered via profiling [66,84], LRU stack hit position counting [156], monitoring [143], or dynamic set sampling [123]. Table 5.1 compares the optimization goals and policies used by prior schemes.…”

Mitigating Memory Resource Contention in Warehouse Scale Computers

“…System software also does not have control over hardware resources such as caches and memory bandwidth, which renders responding to contention and managing applications' QoS quite challenging. As a result, despite the amount of research attention given to contention problems on multicore platforms [28,45,10,19,44,51,52,27,32,15,29,24,63,23,24,7,4,60], mitigating the impact of contention on an application's performance and QoS, enforcing the relative QoS priorities of co-running applications, while maximizing machine utilization, remains key challenges in modern warehouse scale computers.…”

“…Hardware techniques such as cache partitioning and bandwidth partitioning to reduce resource contention and improve performance and fairness on multicores have received much research attention [28,45,10,19,44,51,52,27,32]. In addition, there also have been a number of work aimed at better modeling cache contention [8] and monitoring cache contention [62].…”