Energy-aware load balancing in content delivery networks

Orlando, FL, (2012): 954-962

引用124|浏览108
EI WOS
下载 PDF 全文
引用
微博一下

摘要

Internet-scale distributed systems such as content delivery networks (CDNs) operate hundreds of thousands of servers deployed in thousands of data center locations around the globe. Since the energy costs of operating such a large IT infrastructure are a significant fraction of the total operating costs, we argue for redesigning CDNs to i...更多

代码

数据

0
简介
  • Large Internet-scale distributed systems deploy hundreds of thousands of servers in thousands of data centers around the world.
  • The authors do not assume any particular mechanism, but the authors do assume that those mechanisms allow load to be arbitrarily re-divided and re-distributed among servers, both within a cluster and across clusters
  • This is a good assumption for typical web workloads that form a significant portion of a CDN’s traffic.
  • Based on the own testing of typical off-the-shelf server configurations used by CDNs, the authors use the standard linear model [4] where the power consumed by a server serving load λ is power(λ) =∆ Pidle + (Ppeak − Pidle)λ,
重点内容
  • Large Internet-scale distributed systems deploy hundreds of thousands of servers in thousands of data centers around the world
  • We ran algorithm Hibernate on typical content delivery networks load traces collected over 25 days and across 22 clusters for multiple values of τ and two values of κ with the results summarized in Figure 4
  • We proposed energy optimization techniques to turn off content delivery networks servers during periods of low load while seeking to balance the interplay of three key design objectives: maximize energy reduction, minimize the impact on clientperceived availability (SLAs), and limit the frequency of onoff server transitions to reduce wear-and-tear and its impact on hardware reliability
  • We proposed an optimal offline algorithm and an online algorithm to extract energy savings both at the level of local load balancing within a data center and global load balancing across data centers
  • Our evaluation using real production workload traces from a large commercial content delivery networks showed that it is possible to reduce the energy consumption of a content delivery networks by more than 55% while ensuring a high level of availability that meets customer SLA requirements with only a modest number of on-off transitions per server per day
  • Our future work will focus on the incorporation of workload prediction techniques into our Hibernate algorithm, further optimizations of the global load balancing algorithm from an energy perspective and techniques for managing footprint of content delivery networks customers while turning servers on and off
结果
  • Viewing all the clusters of the CDN as a single system, the system-wide energy reduction by using OPT in all the clusters was 64.2%
  • This implies that significant gains are possible in the offline scenario by optimally orchestrating the number of live servers in each cluster.
  • Figure 3 shows the optimal system-wide energy reduction for each value of the average transitions that is allowable.
  • These numbers were obtained by running algorithm OPT(k) for all clusters for a range of values of k.
  • The modest decrease in energy reduction may well be worth it for most CDNs, since availability is much higher with 10% spare capacity than with no spare capacity requirement (Figure 4c)
结论
  • The authors proposed energy optimization techniques to turn off CDN servers during periods of low load while seeking to balance the interplay of three key design objectives: maximize energy reduction, minimize the impact on clientperceived availability (SLAs), and limit the frequency of onoff server transitions to reduce wear-and-tear and its impact on hardware reliability.
  • The authors show that keeping even 10% of the servers as hot spares helps absorb load spikes due to global flash crowds with little impact on availability SLAs. The authors' future work will focus on the incorporation of workload prediction techniques into the Hibernate algorithm, further optimizations of the global load balancing algorithm from an energy perspective and techniques for managing footprint of CDN customers while turning servers on and off
相关工作
  • Energy management in data centers has been an active area of research in recent years [6]. Techniques that have been developed in this area include, use of DVFS to reduce energy, use of very low-power servers [3], routing requests to locations with the cheapest energy [15] and dynamically activating and deactivating nodes as demand rises and falls [5], [16], [12]. A key difference between much of this prior work and the current work is our focus on CDNs, with a particular emphasis on the interplay between energy management and the local/global load balancing algorithms in the CDN. We also examine the impact of shutting servers on client SLA as well as the impact of server transitions on wear and tear.

    A recent effort related to our work is [13]. Like us, this paper also presents offline and online algorithms for turning servers on and off in data centers. While [13] targets data center workloads such as clustered mail servers, our focus is on CDN workloads. Further, [13] does not emphasize SLA issues, while in CDNs, SLAs are the most crucial of the three metrics since violations can result in revenue losses. Two recent efforts have considered energy-performance or energy-QoS tradeoff in server farms [9], [10]. Our empirical results also show an energy-SLA tradeoff, and we are primarily concerned with choosing system parameters to obtain five 9s of availability in CDNs.
基金
  • The authors would like to acknowledge the support of NSF awards CNS-0519894, CNS-0916972, and CNS-1117221
引用论文
  • M. Adler, R. Sitaraman, and H. Venkataramani. Algorithms for optimizing bandwidth costs on the internet. In Proceedings of the 1st IEEE Workshop on Hot Topics in Web Systems and Technologies (HOTWEB), pages 1–9, Los Alamitos, CA, USA, November 2006. IEEE Computer Society.
    Google ScholarLocate open access versionFindings
  • H. Amur, J. Cipar, V. Gupta, G.R. Ganger, M.A. Kozuch, and K. Schwan. Robust and flexible power-proportional storage. In Proceedings of the 1st ACM symposium on Cloud computing, pages 217–228. ACM, 2010.
    Google ScholarLocate open access versionFindings
  • D. Anderson, J. Franklin, M. Kaminsky, A. Phanishayee, L. Tan, and V. Vasudevan. Fawn: A fast array of wimpy nodes. In Proceedings of ACM SOSP, October 2009.
    Google ScholarLocate open access versionFindings
  • L.A. Barroso and U. Holzle. The case for energy-proportional computing. Computer, 40(12):33–37, 2007.
    Google ScholarLocate open access versionFindings
  • J. Chase, D. Anderson, P. Thakar, A. Vahdat, and R. Doyle. Managing energy and server resources in hosting centers. In Proceedings of the Eighteenth ACM Symposium on Operating Systems Principles (SOSP), pages 103–116, October 2001.
    Google ScholarLocate open access versionFindings
  • A. Chen, W. Das, A. Qin, A. Sivasubramaniam, Q. Wang, and N. Gautam. Managing server energy and operational costs in hosting centers. In Proceedings of the ACM SIGMETRICS International Conference on Measurement and Modeling of Computer Systems, June 2005.
    Google ScholarLocate open access versionFindings
  • G. Chen, W. He, J. Liu, S. Nath, L. Rigas, L. Xiao, and F. Zhao. Energyaware server provisioning and load dispatching for connection-intensive internet services. In Proceedings of the 5th USENIX Symposium on Networked Systems Design and Implementation, pages 337–350. USENIX Association, 2008.
    Google ScholarLocate open access versionFindings
  • J. Dilley, B. Maggs, J. Parikh, H. Prokop, R. Sitaraman, and B. Weihl. Globally distributed content delivery. Internet Computing, IEEE, 6(5):50–58, 2002.
    Google ScholarLocate open access versionFindings
  • A. Gandhi, V. Gupta, M. Harchol-Balter, and M. Kozuch. Optimality analysis of energy-performance trade-off for server farm management. In Proc. 28th Intl. Symposium on Computer Performance, Modeling, Measurements, and Evaluation (Performance 2010) Namur, Belgium, November 2010.
    Google ScholarLocate open access versionFindings
  • K. Kant, M. Murugan, and D H. C Du. Willow: A control system for energy and thermal adaptive computing. In Procedings of the 25th IEEE IPDPS, 2011.
    Google ScholarLocate open access versionFindings
  • J.G. Koomey. Worldwide electricity used in data centers. Environmental Research Letters, 3, Sept 2008.
    Google ScholarLocate open access versionFindings
  • A. Krioukov, P. Mohan, S. Alspaugh, L. Keys, D. Culler, and R. Katz. Napsac: Design and implementation of a power-proportional web cluster. In Proc. of ACM Sigcomm workshop on Green Networking, August 2010.
    Google ScholarLocate open access versionFindings
  • M. Lin, A. Wierman, L.L.H. Andrew, and E. Thereska. Dynamic rightsizing for power-proportional data centers. Proc. IEEE INFOCOM, Shanghai, China, pages 10–15, 2011.
    Google ScholarLocate open access versionFindings
  • E. Nygren, R.K. Sitaraman, and J. Sun. The Akamai Network: A platform for high-performance Internet applications. ACM SIGOPS Operating Systems Review, 44(3):2–19, 2010.
    Google ScholarLocate open access versionFindings
  • A. Qureshi, R. Weber, H. Balakrishnan, J. Guttag, and B. Maggs. Cutting the electric bill for internet-scale systems. In Proceedings of the ACM SIGCOMM 2009 conference on Data communication, pages 123–134. ACM, 2009.
    Google ScholarLocate open access versionFindings
  • N. Tolia, Z. Wang, M. Marwah, C. Bash, P. Ranganathan, and X. Zhu. Delivering energy proportionality with non energy-proportional systemsoptimizing the ensemble. In Proc of Workshop on Power-aware Computing Systems, San Diego, CA, December 2008.
    Google ScholarLocate open access versionFindings
0
您的评分 :

暂无评分

标签
评论
数据免责声明
页面数据均来自互联网公开来源、合作出版商和通过AI技术自动分析结果,我们不对页面数据的有效性、准确性、正确性、可靠性、完整性和及时性做出任何承诺和保证。若有疑问,可以通过电子邮件方式联系我们:report@aminer.cn