Coordinated VM Resizing and Server Tuning: Throughput, Power Efficiency and Scalability

Abstract

Abstract-Performance control and power management in virtualized machines (VM) are two major research issues in modern data centers. They are challenging due to complexities of hosted Internet applications, high dynamics in workloads and the shared virtualized infrastructure. Obtaining a model among VM capacity, server configuration, performance and power consumption is a very hard problem even for just one application. In this paper, we propose and develop GARL, a genetic algorithm with multi-agent reinforcement learning approach for coordinated VM resizing and server tuning. In GARL, model-independent reinforcement learning agents generate VM capacity and server configuration options and the genetic algorithm evaluates different combinations of those options for maximizing a global utilization function of system throughput and power efficiency. The multi-agent design makes GARL a scalable approach, which is important as more and more applications are hosted in data centers using cloud services. We build a testbed in a prototype data center and deploy multiple RUBiS benchmark applications. We apply a power budget in the testbed and observe superior system throughput and power efficiency of GARL. Experimental results also find that GARL significantly outperforms a representative reinforcement learning based approach in performance control. GARL shows better scalability when compared to a centralized approach.

pVOCL: Power-Aware Dynamic Placement and Migration in Virtualized GPU Environments

“…In contrast, pVOCL dynamically adjusts the task across different GPUs via live task migration and reflect real-time state of the system. There are significant research efforts focused on power, energy and thermal management in enterprise data centers [3]- [5], [7]- [9], [12]- [14], [16]. However, existing techniques are not easily applicable to GPU-accelerated high performance computing systems.…”

TARGO: Transition and Reallocation Based Green Optimization for Cloud VMs

“…EnaCloud [9] comprises two key components: Global Controller which manages job scheduling and distribution tasks, and Resource pool where VM hypervisors perform monitor, sizing, and power management control. GARL [24], on the other hand, is equipped with VM Power Monitor and VM Resource Allocator which facilitate VM resizing and replacement according to VM performance monitor data. The authors of [22] utilize workload forecasting statistics as the basis to facilitate VM sizing and resizing consolidation tasks.…”

“…In this context, much GCC research has been done addressing on virtualization and VM management optimizations. Current approaches to achieve better energy and (computational) resource utilization efficiency can be generally classified into two categories: 1) server/workload consolidation solutions involving VM scheduling [1][14] [16], migration [15][18] [26], and sizing/resizing management [9][22] [24]; 2) hardware integrated solutions involving dynamic voltage and frequency scaling (DVFS) [25] and hardware cooling control mechanism [12]. As a matter of fact, these approaches can only be implemented on Cloud service provider (CSP) side (inside the Cloud).…”

Self-Tuning Batching with DVFS for Improving Performance and Energy Efficiency in Servers

“…There are recent studies on coordinated power and performance management in virtualized servers [1], [6], [9], [11], [14], [12], [15], [19], [28]. Mistral [11] is a control architecture for optimizing power consumption, performance benefit, and the transient costs in Cloud environments.…”