A Reinforcement Learning Approach to Online Web Systems Auto-configuration

Bu, Xiangping; Rao, Jia; Xu, Cheng‐Zhong

doi:10.1109/icdcs.2009.76

Cited by 86 publications

(61 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We refer to this agent as App-Agent. It monitors the performance of appliances belonging to the same application and refine their configurations through interactions with the appliances in a way similar to VM-Agent to meet the application's SLO requirement [5]. Unlike VM-Agent, App-Agent is run in a separate VM, requiring no OS or hardware level information.…”

Section: Url Framework For Autoconfigurationmentioning

confidence: 99%

URL: A unified reinforcement learning approach for autonomic cloud management

Rao

2012

Journal of Parallel and Distributed Computing

Self Cite

137

View full text Add to dashboard Cite

Section: Url Framework For Autoconfigurationmentioning

confidence: 99%

URL: A unified reinforcement learning approach for autonomic cloud management

Rao

2012

Journal of Parallel and Distributed Computing

Self Cite

137

View full text Add to dashboard Cite

“…A few recent studies focused on automated server parameter tuning for multi-tier server systems [2], [7]. In our recent study [7], we proposed to use the effective system throughput as the primary performance metric for multi-tier Internet applications.…”

Section: Related Workmentioning

confidence: 99%

“…As many others in [2], [11], [12], [20], [24], [30], we use RUBiS [1] as the benchmark application in conducting the experiments. RUBiS is an open source multi-tier Internet benchmark application.…”

Section: B the Genetic Algorithm In Garl For Coordinationmentioning

confidence: 99%

See 1 more Smart Citation

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

Guo

Zhou

2012

2012 IEEE 20th International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems

View full text Add to dashboard Cite

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.

show abstract

“…Existing solutions are typically based on using queuing theory to construct analytical performance models for applications executed on VMs (Doyle et al 2003, Bennani andMenascé 2005), applying control theory to automatically adjust VM resource allocation and achieve the desired application performance , (Wang, Zhu, and Singhal 2005), (Padala et al 2008), and using machine learning techniques to automatically learn the resource model for a virtualized system based on data observed from the system (Wildstrom, Stone, and Witchel 2008), (Wood et al 2008), (Bu, Rao, and Xu 2009), (Xu et al 2008), (Kundu et al 2010). …”

Section: Virtual Machine Resource Managementmentioning

confidence: 99%

Model-driven network emulation with virtual time machine

Liu¹,

Rangaswami²,

Zhao³

2010

Proceedings of the 2010 Winter Simulation Conference

View full text Add to dashboard Cite

We present VENICE, a project that aims at developing a high-fidelity, high-performance, and highly-controllable experimental platform on commodity computing infrastructure to facilitate innovation in existing and futuristic network systems. VENICE employs a novel model-driven network emulation approach that combines simulation of large-scale network models and virtual-machine-based emulation of real distributed applications. To accurately emulate the target system and meet the computation and communication requirements of its individual elements, VENICE adopts a holistic machine and network virtualization technique, called virtual time machine, in which the time advancement of simulated and emulated components are regulated in complete transparency to the test applications. In this paper, we outline the challenges and solutions to realizing the vision of VENICE. A HARMLESS NECESSARY CATWe classify existing experimental networking research testbeds into physical, emulation, and simulation testbeds. Emulation testbeds combine physical testbeds and emulation, and provide the ability to conduct extensive live experiments. Although emulation allows a certain level of flexibility, the emulated network conditions are inherently restricted by physical limitations (such as nodal processing speed, link bandwidth and latency) and cannot be scaled up beyond the capabilities of the underlying physical infrastructure. Comparatively, network simulation is effective at capturing high-level design issues, answering what-if questions, and investigating complex system behaviors, such as multiscale interactions, self-organizing characteristics, and emergent phenomena. However, simulation fairs poorly in other aspects, particularly in its operational realism. Further, simulation development is both labor-intensive and error-prone; reproducing realistic network topology, traffic and diverse operational conditions in simulation is known to be a substantial undertaking.In this position paper, we describe our ongoing efforts on the VENICE project (Virtual Emulation Network Infrastructure for Commodity Environments), which aims to combine the advantages of simulation and emulation techniques to provide a high-fidelity, high-performance, and highly-controllable experimental platform for large-scale network experiments. The central thrust of this project is a novel model-driven network emulation approach for conducting large-scale network experiments. We pursue holistic machine and network virtualization techniques that allow execution of network simulation models and emulation of real implementations of network protocols and distributed applications under a coordinated virtual time-line. VENICE consists of simulated networks connecting to emulated hosts sandboxed inside virtual machines running real implementations of network applications or network protocols (e.g., router software). Traffic generated from the applications is conducted by the virtual network with appropriate delays and losses according to both simulated and emulat...

show abstract

A Reinforcement Learning Approach to Online Web Systems Auto-configuration

Cited by 86 publications

References 10 publications

URL: A unified reinforcement learning approach for autonomic cloud management

URL: A unified reinforcement learning approach for autonomic cloud management

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

Model-driven network emulation with virtual time machine

Contact Info

Product

Resources

About