Aaron Baldwin
Information Technology Department
George Mason University

Networks and the Internet are accessible across majority of the modern world. These massive webs of communication are supported by a conglomerate of information hoarding datacenters, which store the desired information of consumers. These datacenters are essentially a central location for groupings of interlinked servers. Common networks are supported by a mesh of thousands of these datacenters, containing an estimated forty-four million servers; the Internet powerhouse Google accounts for nearly 2% of the world’s active servers (Nielson 2009). Not only does a standard server account for an average power consumption of between 250 to 500 watts per hour, these machines generally run on an uninterrupted cycle for the majority of their lives (West 2010). Through simple mathematics, you can easily find that the low-end estimates of global power consumption by servers accounts for energy usage of over 250 billion watts annually, or 250 GW-hours. These constantly running machines, when looking at the low estimate of usage, account for roughly the annual energy consumption of Greenland (Central 2009). For devices that account for such a significant proportion of the world’s power consumption, it would seem these machines would be strictly regulated for efficiency, this if far from the case. These millions of devices are performing energy and resource management at a level which would render any other machine unusable.

Servers run uninterrupted for the majority of their energy consuming lives, yet they generally remain in an idle state. Unless a server has a client request for its services, it patiently waits until it is called upon. Due to this, servers are generally put to use only 5-15% of the time they are in operation (Kumar 2010). A server’s utilization of an idle state gives the misconception that the machine is consuming the bare minimum of resources; this could not be farther from the truth. While the machines do minutely reduce their power consumption while idle, they still absorb 60 to 90% of their normal workload’s resources (Kumar 2010). Sadly, this blatant misuse of energy is merely half the issue with server inefficiency. As these devices are constantly burning off our highly valuable energy resources, each machine is not even scraping the surface of its storage capabilities. Of the tens of millions of servers in operation, each is leaving 90-95% of their server capacity unused. This means we are inefficiently running ten to twenty times more servers than we need, while each machine siphons our power grid for an absurd amount of energy it does not even use (Chu 2008). The combination of these server’s efficiency issues results in each machine emitting four tons of CO2 annually each year, they are pumping 175 million metric tons of CO2 into the atmosphere annually; more than Thailand’s overall total (Moore 2003). With our environment screaming for some kind of consolidation, a new breed of server has finally begun implementation. An industry-wide push has targeted all of our above listed issues through the newly designed virtualized servers.

Summary of server issues

• Uninterrupted power consumption
• Only put to use 5-15% of their lives
• Consume 60-90% of normal consumption, even while idle
• 90-95% of server capacity goes unused