Wednesday , June 20 2018

Dynamic Server Provisioning for Energy Efficient Large Scale Video-on-Demand Systems

Jian YANG
School of Information Science and Technology, University of Science and Technology of China (USTC)
Hefei, Anhui 230027, China

Ke ZENG
School of Information Science and Technology, University of Science and Technology of China (USTC)
Hefei, Anhui 230027, China

Abstract: In this paper, we propose a server provisioning strategy for energy conservation in large scale VOD systems, which dynamically turns on/off servers in order to adaptively tailor active servers to dynamic user load. By defining quality-of-service (QoS) in terms of system overload probability, we reinterpret the energy conservation problem into minimizing the number of active servers subject to a QoS requirement. Relying on designing a recursive least square (RLS) based online user number predictor and constructing a large deviation based overload probability estimation model, we derive a practical dynamic server provisioning strategy which does not require any prior knowledge about the future workload. Finally, the experiments are carried out based on synthetic and real workload respectively to investigate the achievable performance of the proposed strategy.

Keywords: Video-on-demand, server energy conservation, quality-of-service, load migration, request dispatching.

>>Full text
CITE THIS PAPER AS:
Jian YANG, Ke ZENG, Dynamic Server Provisioning for Energy Efficient Large Scale Video-on-Demand Systems, Studies in Informatics and Control, ISSN 1220-1766, vol. 20 (4), pp. 355-362, 2011.

1. Introduction

Streaming media is undergoing a dramatic growth fuelled by a variety of applications such as internet protocol television (IPTV) and Telco video. Naturally, video-on-demand (VOD) service has to face a challenge of the tremendous concurrent users increase. In order to satisfy the increasingly demanding traffic and the growing consumer population, most VOD systems are deployed in a form of massive server clusters for providing low cost, high performance, availability and scalability. However, the servers today consume ten times more power than they did ten years ago [1], which implies that the large scale server clusters may result in a high power consumption. In this paper, we consider the problem of energy conservation in large scale VOD server clusters.

In recent years, the problem of energy conservation for server clusters has received an increasing attention from both academia and industry. Bianchini et al. [2] gave an overview of different power and energy management techniques of server systems. Gandhi et al. [3] considered the problem of allocating an available power budget among servers to minimize mean response time. Chen et al. [4] presented a solution to the problem of reducing server energy cost at hosting center running multiple applications towards the goal of meeting performance based on Service Level Agreements (SLA). Qureshi et al. [5] characterized the variation due to fluctuating electricity prices and stated that existing distributed systems should be able to exploit this variation for significant economic gains.

Chase et al. [6] studied dynamic provisioning for web cluster to improve the energy efficiency. Chen et al. [7] proposed power saving techniques for connection-intensive services, and evaluated the techniques by using data traces from Windows Live Messenger. Other works [8-11] also studied dynamic provisioning and load dispatching algorithms for web clusters or data centers to improve the energy efficiency.

It should be noted that the most prior works focus on generic requests of web services, rather than multimedia jobs. To the best of our knowledge, energy saving in large scale VOD systems has not been covered yet. Our works for power-efficient large scale VOD server clusters are motivated by two observations.

First, the number of users varies largely during a day [13], which is based on a large VOD system deployed by China Telecom. Second, as shown in [3, 7], an idle server may consume around 60% of the peak power, because the power required to run the OS and hardware is not ignorable. These observations imply that we can achieve substantial energy saving by adjusting the active server according to the time-varying workload, especially shutting down idle servers during off-peak hours.

In this paper, we design a recursive least squares (RLS) based predictor to estimate the forthcoming number of users corresponding to each video, which assists our strategy to determine the server provisioning for the near instantaneous workload without any prior statistical knowledge. In order to conceive a QoS provisioning energy conservation strategy for large scale VOD systems, we define the QoS requirement in terms of the overload probability. Then, we apply large deviation theory [14] to estimate the overload probability that the bandwidth provided by the current active servers cannot satisfy the required bandwidth for the predicted workload. Finally, we use the overload probability to derive an adaptive strategy to find the minimum cluster scale satisfying the QoS requirement.

REFERENCES:

  1. U.S. Environmental Protection Agency, Server and Data Center Energy, 2007.
  2. BIANCHINI, R., RAJAMONY, R., Power and Energy Management for Server Systems, IEEE Computer, vol. 37, 2004, pp. 69-76.
  3. GANDHI, A., HARCHOL-BALTER, M., et al., Optimal Power Allocation in Server Farms, Proc. ACM SIGMETRICS/ Performance, 2009.
  4. CHEN, Y., DAS, A., et al., Managing Server Energy and Operational Costs in Hosting Centers, Proc. ACM SIGMETRICS, 2005.
  5. QURESHI, A., WEBER, R., et al., Cutting the Electric Bill for Internet-Scale Systems, Proc. ACM SIGCOMM, 2009.
  6. CHASE, J.S., ANDERSON, D.C., et al., Managing Energy and Server Resources in Hosting Centers, Proc. ACM SOSP, 2001.
  7. CHEN, G., HE, W., et al., Energy-Aware Server Provisioning and Load Dispatching for Connection-Intensive Internet Services, Proc. USENIXNSDI, 2008.
  8. JOUKOV, N., SIPEK, J., GreenFS: Making Enterprise Computers Greener by Protecting Them Better, in Proc. ACM EuroSys, 2008.
  9. PADALA, P., HOU, K., et. al., Automated Control of Multiple Virtualized Resources, Proc. ACM Eurosys, 2009.
  10. GANDHI, A., GUPTA, V., et al., Optimality Analysis of Energy-Performance Trade-offfor Server Farm Management, Proc. IFIP Performance, 2010.
  11. URGAONKAR, R., KOZAT, U.C., et al., Dynamic Resource Allocation and Power Management in Virtualized Data Centers, Proc. IEEE/IFIPNOMS, 2010.
  12. http://traces.cs.umass.edu/index.php/network /Network.
  13. YU, H., ZHENG, D., et al., Understanding User Behavior in Large-Scale Video-on-Demand Systems, Proc. ACM EuroSys, 2006.
  14. BILLINGSLEY, P., Probability and Measure, 2nd ed., Wiley, NewYork, 1986.
  15. KELLY, F.P., Effective bandwidths at multi-class queues, Queueing Systems: Theory and Applications, vol.9, 2001, pp.5-16.
  16. HAYKIN, S., Adaptive Filter Theory Third Edition. Upper Saddle River, NJ: Prentice-Hall, 1996.
  17. http://trace.eas.asu.edu/mpeg4/index.html.
  18. ALMEIDA, J.M., KRUEGER, J., et al., Analysis of educationa lmedia server workloads, Proc. ACM NOSSDAV, 2001.
  19. CHERKASOVA, L., GUPTA, M., Analysis of Enterprise Media Server Workloads: Access Patterns, Locality, Content Evolution, and Rates of Change, IEEE/ACM Trans. Netw., vol. 12, 2004, pp. 781-794.
  20. http://www.kasenna.com/kasenna/static/products/hardware.html.

https://doi.org/10.24846/v20i4y201103