CoPart: Coordinated Partitioning of Last-Level Cache and Memory Bandwidth for Fairness-Aware Workload Consolidation on Commodity Servers

Abstract

Workload consolidation is a widely-used technique to maximize server resource utilization in cloud and datacenter computing. Recent commodity CPUs support last-level cache (LLC) and memory bandwidth partitioning functionalities that can be used to ensure the fairness of the consolidated workloads. While prior work has proposed a variety of resource partitioning techniques, it still remains unexplored to characterize the impact of LLC and memory bandwidth partitioning on the fairness of the consolidated workloads and investigate system software support to dynamically control LLC and memory bandwidth partitioning in a coordinated manner.

To bridge this gap, we present an in-depth performance and fairness characterization of LLC and memory bandwidth partitioning. Guided by the characterization results, we propose CoPart, coordinated partitioning of LLC and memory bandwidth for fairness-aware workload consolidation on commodity servers. CoPart dynamically analyzes the characteristics of the consolidated applications and allocates the LLC and memory bandwidth across the applications in a coordinated manner to improve the overall fairness. Our quantitative evaluation shows that CoPart significantly improves the fairness of the consolidated applications (e.g., 57.3% higher fairness on average than the resource allocation policy that equally allocates the resources to the consolidated applications), robustly provides high fairness across various application and system configurations, and incurs small performance overhead.