One topic which will be given some prominence in certain chapters of our book is that of energy hungry IT hardware. Just like the Cookie Monster, typical servers today have an insatiable appetite. Instead of craving cookies however, servers crave electricity. Two obvious direct by-products of their electricity consumption is both CO2 and heat.
The degree to which this craving changes as the hardware workload changes is referred to as its ‘energy elasticity’. For example, a server which has a high degree of energy elasticity is one characterised by consuming significantly less power when it’s idle compared to when it’s running at its maximum processing potential. Conversely, a server which has a low degree of energy elasticity consumes almost the same amount of electricity whether it’s in use or idle. Over the years the energy elasticity of commercial servers, desktops and laptops alike has steadily improved as hardware manufacturers have become more environmentally conscience (driven by market forces of course). The result is that the speed of adoption of energy elasticity improving technology has hastened, but there is evidence that significantly more can be done. Google for example uses servers which are “commodity-class x86 PCs running customised versions of Linux” which according to a paper published in the MIT Technical Review entitled “Energy-Aware Internet Routing” consume an impressive 65 percent of their normal power when idle. These servers clearly possess a high degree of energy elasticity. Google has also been able to significantly reduce the amount of heat emitted by their customised servers. These points alone suggest that it is possible to make commodity-class servers far more energy elastic with only simple, intelligent modifications.
Just like Google’s customised servers, there are numerous other interesting hardware and software solutions to the energy elasticity problem. However, there does seem to be a preference by these same solution providers for focusing on efficiency rather than elasticity. My take on this is due to the fact that historically the focus has been on squeezing the most performance out of the hardware and not its environmental footprint. There has been no reason for solution providers to service a market that jut doesn’t exist.
One example on the software side of the efficiency argument is a new linux distribution that gained some well deserved media attention this week called Lubuntu linux distribution This distribution is a derivative of Ubuntu with a light weight desktop called LXDE (Lightweight X11 Desktop Environment) and aimed at running on lower spec’ed machines. These machines typically use much less power (and in turn are responsible for lower CO2 emissions) than their more power counterparts. Running an operating system like Lubuntu on a fit-PC2 would consume a mere 5 watts of electricity per hour.
So is it better to look at efficiency or elasticity? I would say they are two very important components of the greener machine solution. Putting numbers to how large the elasticity component would however be very difficult to calculate accurately, and is only made more difficult by the fact that quality data on even the number and type of servers world wide is sparse. This very point is made by Jonathan G. Koomey, Ph.D. (a staff scientist at Lawrence Berkeley National Laboratory and a consulting professor at Stanford University, who studies information technology energy use and environmental impact) who points out that high quality research in this area is hard to come by due to the proprietary and secretive nature of enterprises. This is absolutely understandable – IT is still today a strategic asset (well, Nicholas Carr might disagree).
According to IDC, a server (clearly a very general reference) operating at 10% utilisation still consumes the same power and cooling as a server operating at 75% utilisation. They claim that in the current state of the market, for every $1.00 spent on new servers, the average enterprise spends $0.50 on power and cooling. Therefore energy elasticity is clearly a large and substantial part of the green server solution.
Ignoring the actual numbers though, consider three scenarios:
- Enterprise X has a large server farm which is 60% utilised during a typical business day and during weekends and evenings is 5% utilised. This enterprise keeps all its servers running all the time to minimise operational/maintenance costs and to ensure power is available whenever its needed – day or night.
- Enterprise Y also has a large server farm with similar demand patterns. However, in an attempt to introduce energy elasticity into the enterprise further than what’s build into the hardware, Enterprise Y runs on a virtualised environment. As demand varies virtual machines are consolidated, and unused servers are turned on/off (automatically) as demand dictates.
- Enterprise Z has no processing capabilities itself, but instead leverages a cloud computing provider like GoGrid or Amazon. Their entire architecture is based around spawning and destroying instances of virtual servers in the cloud as demand fluctuates.
The energy elasticity of Enterprise X’s IT infrastructure is limited to what the hardware and software itself can provide. Enterprise Y in comparison has the opportunity to dramatically reduce its power usage thanks to virtualisation. However there is inevitably a constant (albeit far smaller than Enterprise X’s) power footprint. Finally Enterprise Z has an ultimately elastic architecture but without knowing how the cloud computing provider’s underlying infrastructure has been implemented it may or may not be the best. The energy elasticity enjoyed by Enterprise Y and Enterprise Z will in all likelihood result in cost and CO2 savings that surpass even the most elastic individual servers that Enterprise X may employ. In todays world of distributed computing I would suggest that the biggest gains in energy elasticity are not the concern of hardware vendors but are instead the concern of application and operational architects. These architects are positioned correctly to build systems that have energy elasticity built into them while at the same time providing the customer with solutions to their business needs.
If you have an opinion on this topic or have any material that might be of interest to the book we are writing on Green IT please leave a comment to this post.