Greening web servers: A case for ultra low-power web servers

This paper studies the feasibility and benefits of greening Web servers by using ultra-low-power micro-computing boards to serve Web content. Our study focuses on the tradeoff between power and performance in such systems. Our premise is that low-power computing platforms can provide adequate performance for low-volume Websites run by small businesses or groups, while delivering a significantly higher request per Watt. We use the popular Raspberry Pi platform as an example low-power computing platform and experimentally evaluate our hypothesis for static and dynamic Web content served using this platform. Our result show that this platform can provide comparable response times to more capable server-class machines for rates up to 200 requests per second (rps); however, the scalability of the system is reduced to 20 rps for serving more compute-intensive dynamic content. Next, we study the feasibility of using clusters of low-power systems to serve requests for larger Websites. We find that, by utilising low-power multi-server clusters, we can achieve 17x to 23x more requests per Watt than typical tower server systems. Using simulations driven by parameters obtained from our real-world experiments, we also study dynamic multi-server policies that consider the tradeoff between power savings and overhead cost of turning servers on and off.

[1]  Virgílio A. F. Almeida,et al.  Measuring the behaviour of a world-wide web server , 1997, HPN.

[2]  David Mosberger,et al.  httperf—a tool for measuring web server performance , 1998, PERV.

[3]  Laurent Massoulié,et al.  Greening the internet with nano data centers , 2009, CoNEXT '09.

[4]  Carey L. Williamson,et al.  Internet Web servers: workload characterization and performance implications , 1997, TNET.

[5]  Guangwei Bai,et al.  Performance benchmarking of wireless Web servers , 2007, Ad Hoc Networks.

[6]  C. Amza,et al.  Specification and implementation of dynamic Web site benchmarks , 2002, 2002 IEEE International Workshop on Workload Characterization.

[7]  Erez Zadok,et al.  Optimizing energy and performance for server-class file system workloads , 2010, TOS.

[8]  Niklas Carlsson,et al.  Improving the scalability of a multi-core web server , 2013, ICPE '13.

[9]  Daniel A. Menascé,et al.  TPC-W: A Benchmark for E-Commerce , 2002, IEEE Internet Comput..

[10]  Amar Phanishayee,et al.  FAWN: a fast array of wimpy nodes , 2009, SOSP '09.

[11]  Mark Teel Using wview , 2010 .

[12]  Prashant J. Shenoy,et al.  Energy-aware load balancing in content delivery networks , 2011, 2012 Proceedings IEEE INFOCOM.

[13]  Luiz André Barroso,et al.  The Case for Energy-Proportional Computing , 2007, Computer.

[14]  Niklas Carlsson,et al.  Helping Hand or Hidden Hurdle: Proxy-Assisted HTTP-Based Adaptive Streaming Performance , 2013, 2013 IEEE 21st International Symposium on Modelling, Analysis and Simulation of Computer and Telecommunication Systems.

[15]  Werner Vogels,et al.  Beyond Server Consolidation , 2008, ACM Queue.