The energy-efficiency of server hardware, web server software, and databases has been widely studied. However, studies that combine these aspects are rare. In this chapter, the authors present an energy-efficiency evaluation of a web/database application in a Windows/IIS/MSSQL environment running on an industrial grade Intel server. Moreover, they provide a wide overview of related research and technologies. Researchers have noticed that despite energy-saving technologies, energy consumption of data centers is still growing. To resolve this dilemma, the authors explore the background and propose concrete solutions. They concentrate on the following aspects: server BIOS/operating system energy optimization (limited impact) and “bursting” (i.e., queuing requests and then executing them in bursts). The authors have used the bursting method with both database and web/database applications. Their results indicate about 10% energy savings using this method. The authors analyse the model using statistical tools and present an equation to express the quality of service vs. burst wait time relationship.
TopIntroduction
Rapidly increasing energy consumption of the IT sector is a widely recognised issue because of its large economic and environmental impact. Patel and Shah estimate (Patel and Shah, 2005) the burdened cost of power and cooling, inclusive of redundancy, to be 25% to 30% of the total cost of ownership in typical enterprise data centers. At the same time, CO2 emissions of the IT sector are on the same level as emissions of the air traffic. In their 2014 Digital Universe Study the International Data Corporation evaluated the amount of information created or replaced to be 4.4 trillion gigabytes in 2013 and growing by a factor of 10. The IT energy consumption is assumed to grow just as rapidly (Li et al., 2011).
The problem has been noticed in the scientific world. Green computing, as it is now often referred to, seeks to improve the energy efficiency of computing centers. This is a wide topic incorporating issues like locating data centers near cheap energy sources (Brown and Reams, 2010), minimizing so called e-waste (Hanselman and Pegah, 2007), designing optimal cooling infrastructure and running the center in an optimal way (Marwah et al., 2009).
Our work has been mainly concerned with software aspects of green computing as with virtualization (Kommeri et al., 2012) and scheduling (Niemi et al., 2012). In this study we concentrate on more specific aspects: (i) studying the energy efficiency of different software components of web-based applications as (Economou et al., 2006), and (ii) illustrating how energy consumption can be decreased by increasing the idle time of the database and web servers. Since the current hardware is not energy proportional (Barroso and Hölzle, 2007), i.e, the energy consumption of the system is not a linear function of the load, rather it is usually more efficient to run the system near full power or let it idle. Our results show that this method can offer over 10% energy savings, but the trade-off is a longer response time. However, we provide a model and equation to analyze the balance of the energy saving and response time.
Our case study focuses on the power consumption of a web/database system that emulates an actual application. The content of the database corresponds with the High Energy Physics Inspire catalog1 maintained at CERN, at the time when the catalog contained ca 800 000 bibliographic records.
With our tests, we want to study the following facets of energy consumption in a modern multi-core web/database system:
Q1: What are the proportions of the database component and the web server component in the total energy consumption?
Q2: How does the computer’s energy-saving configuration (operating system settings vs. BIOS settings) affect the energy consumption?
Q3: Does running server requests as bursts (sequence of requests, then a pause) improve energy efficiency compared to running them linearly?
Our main contributions are (i) evaluating both web application and database performance and energy efficiency (ii) developing an actual database/web application with real data (iii) using both synthetic data, real data and actual web log requests in our tests and (iv) developing a model for practical evaluation of the energy vs response time by statistical tools. We note, too, that simply turning off cores in a multi-core computer does not necessarily save energy.
This chapter is an updated version of our previous article (Kommeri et al. 2014). The rest of the chapter is organized as follows. After a survey of related research in Background, we describe our case application in Section Web/Database Application. In Section Methodology we present our testing methodology and the test environment (server hardware, software, energy meter). In Section Results we present the results. The last section contains the conclusions and future work.