Resilient Workload Manager: Taming Bursty Workload of Scaling Internet Applications

摘要

In data centers hosting scaling Internet applications, operators face the tradeoff dilemma between resource efficiency and Quality of Service (QoS), and the root cause lies in workload dynamics. In this paper, we address the problem with the design of Resilient Workload Manager (ROM). ROM explicitly segregates base workload and trespassing workload, the two naturally different components in application workload, and manages them separately in two resource zones with specialized optimization techniques.rnAs a comprehensive workload management framework, ROM covers workload, data, resource, and Quality of Service of the target applications. It features a fast workload factoring algorithm for distributing incoming application requests, not only on volume but also on content, between the two resource zones; integrated two-dimensional workload shaping, resource planning, and request dispatching schemes for efficient utilization of base workload zone resource; and a simple and high-performance system architecture for dynamic provisioning in trespassing workload zone. Through extensive evaluation, we showed ROM can achieve resource efficiency (e.g., 54.9% server saving), guarantee QoS (based on client-side perceived service quality), reduce data access overhead in the trespassing workload zone during peak load (up to two orders of magnitude), and be adaptive at processing speed (running faster at peak load periods than at regular periods).