A large problem in many data centres is the typical dated view, that one application belongs on one server. This problem leads to server sprawl, which means servers are underutilised. In other words they are not being used to their full potential and are inefficient for the amount resources they are consuming (Rouse, 2008).
Furthermore to illustrate the severity of underutilisation Barouldi et al. (2009) state that levels of server utilisation are thought to be typically between 10-20% per server, which means there is 80%+ of CPU capacity which is going unused. Levels of server utilisation make little difference to the server’s power consumption. Pretorius et al. (2011) state that even at low CPU utilisation such as 10%, servers are still consuming up to 90% of their maximum power consumption. This is an obvious demonstration of power wastage. In order to combat server sprawl, a technology known as Virtualisation can be adopted. Virtualisation is the process of running multiple operating systems on the same underlying piece of hardware(VMware Inc, 2006). Virtualisation allows you to increase the utilisation of the CPU in the physical server from between 15- 85%(Golden, 2010). The process of moving several servers onto one physical server is known as server consolidation (Torres et al., 2008). Server Consolidation will offer large amounts of saving on power consumption, because if several physical servers are virtualised and merged onto one physical server. This means that there will be fewer physical servers which reduces power consumption, cooling requirements and increases rack space and server utilisation. (Torres et al., 2008) Research has outlined various benefits from using virtualisation. O’Neill (2010) states, virtualisation will increase the ease of migrating servers to and from different machines. This is because all an administrator has to do is move the VMimage between servers (or use clustering/VMotion etc for automation) and it will work flawlessly. This will reduce any likelihood of down time especially if the host server is to undergo maintenance. He also states that virtualisation allows for dynamic resource allocation rather than fixed. This means should one of the virtual servers require additional RAM, the HyperVisor (VM handler used to control virtual machines) can deliver this. Similarly if a server is not using its allocated RAM, this can be dynamically allocated to another guest server operating system. Additionally virtualisation offers quick and fast disaster recovery so rather than it taking a few days to recover a server, you can simply recreate the server using snapshot features. There have been numerous case studies carried out which test the effectiveness of virtualisation as a way of increasing efficiency. Simpson (2007) carried out a case study on Australian company, Fosters. She states that Fosters had 30 physical servers in their European data centre. This was reduced to 4 physical servers and 15 virtualised servers. She found there to be a large reduction in power usage thus leading to large savings. Pretorius et al. (2011) carried out a pilot study at Greenwich university. This involved modelling a virtual infrastructure and comparing it to its physical counterpart. They found that by virtualising 12 out of the 21 servers, carbon was reduced by up to 31% when compared to the physical counterparts.
(References available on request)
Greening the Data Centre Project I have managed and had pleasure to work with Auzzam Chaudhri