in this paper, scientists from state university of ny, binghampton, who are also a part of the national science foundation es2 lab address the problem of improving the energy efficiency of servers that provide web-based services, including services provided through clouds.

in recent years there has been an increased demand for web based applications such as media streaming, search services, e commerce, email, social networking, and cloud computing.  the growth in these services has been fuelled in part with the widespread use of web-enabled mobile devices. such unprecedented growth in the demand for these services has spurred a growth in the number and capacities of data centers worldwide. data centers are a growing source of the total national energy consumption: based on 2005 data [1], the total energy consumption of all u.s. datacenters alone was about 1.5 percent of the u.s. energy consumption. since 2005, the energy consumed by data centers has also grown annually in double digit percentages, demanding the need to design and operate data centers in an energy efficient manner. this paper addresses a central issue in reducing the energy consumption of data centers. we propose and evaluate a technique to operate servers in an energy-efficient manner. the proposed scheme recognizes the low energy efficiency of modern servers at light load levels [2,3] and operates the fewest number of servers that are needed to cope with the current load level. the active servers are operated at a high loading level to realize a high degree of energy efficiency (as measured by performance per joule of energy spent). inactive servers are maintained in the deep sleep mode whenever possible to reduce energy wastage from idling. current system-level load trends are used to turn servers on in advance to deal with increase in load levels. energy measurements from a proto type implementation demonstrate that significant energy savings can be realized over a baseline load directing scheme within typical operating regions of a traditional server installation.

the technique proposed and evaluated here has the following characteristics:

• energy-aware load direction: it uses realistic, direct measures of server loading for directing the offered workload to servers.

• request-oblivious redirection: it performs load direction without any a priori knowledge of the individual requests.

• simplicity: it uses a simple control mechanism to steer workload towards active servers and to activate/deactivate servers in anticipation of an increase or reduction in the offered workload.

• independence: the load direction scheme permits each server to use any local energy management scheme as they see fit.

click here to read the full pdf article at the university of new york at binghampton web site, there is no cost for this article.

More >>