Energy-Efficient Monitoring and Controlling of Computer Systems

Background

As described in the introduction, the increasing worldwide demand on energy for data centres is becoming a serious problem. If the energy required running modern data centres would increase at a constant rate over the next years, we would theoretically need all power plants of the world just to run the data centres in a few years. For this reason, the European Council has confirmed the objective to save 20% of the EU's energy consumption compared to projections for 2020 (CEC, 2009), while the Climate Group (2008) has estimated that ICT-enabled improvements could save about 15% of total carbon emissions by 2020, even in other sectors than the ICT sector. Every IT company must deal with this emerging problem that more and more companies are already aware of. By now there are several rather conventional factors that can increase the energy efficiency of data centres, which will be briefly presented in the following passage.

The first possibility to increase energy efficiency is to scrutinise existing data and applications. Often it is possible to shut down some servers just because the running application or existing data is no longer used, but in many scenarios the necessity of applications is not monitored, which makes it impossible to evaluate the necessity of servers. Also, the amount of resources used by different programs might vary a lot and should be compared critically.

The second factor is to use server virtualisation. By doing so it is possible to migrate old servers to virtual machines (VM) and to consolidate several VMs on one physical server (host). In most cases, a prerequisite is a newer hardware with CPU support for Virtualisation Technology (called Intel VT or AMD-V, see details at AMD (2005)). Customers reported an energy drop of up to 80% in BMU (2009) while increasing the utilization of the hosts from 5-10% to 60-80%.