Code Green
There is a huge focus in the media today relating to carbon emissions. The auto industry, rightly so is being forced to dramatically reduce the levels of pollution and carbon emissions generated through transport. Countries are committed to creating zero carbon energy sources by the mid-century. So what is happening within the Computer industry.
An initial review will indicate that we are doing quite a lot, data centres are becoming more energy efficient, cloud service providers are maximising computer use and doing away with idle wasteful CPU cycles, the next-generation of networks are delivering much greater bandwidth with a smaller carbon footprint. So yes things appear to be going well, but the problem appears to be that emissions from compute show no sign of reducing and are in fact rapidly rising. Why is this? The first obvious factor is the digitalisation of everything. We are moving towards a more digital world and as a consequence we are going to need more compute. Just as the auto industry is expected to deliver cars which produce less emissions per kilometre travelled, it should be expected that application developers provide functions which require less energy to execute. Unfortunately, this is not the case, in fact superefficient cloud data centres are having the opposite effect on the energy efficiency of code.
I may need to go back a little bit to the early days of computing. When software development really got started compute was astronomically expensive, so developers used ingenious ways to deliver their functions. Now the cost of compute is so low that the focus is entirely around rapid delivery of functionality and there is little incentive for developers to deliver energy efficient software. I would go one step further to say that a developer using cloud infrastructure doesn’t know how much energy the software might be consuming.
So what can be done? Just like the auto industry when left to a free market model the incentive of energy efficiency is based on the cost of fuel and consumer behaviour. We saw that as fuel as a proportion of disposable income dropped, people no longer saw it as important and bought bigger less energy-efficient vehicles. With compute the same problem exists, energy is such a small proportion of the total cost of application delivery increases in energy costs have little effect on consumer behaviour.
The true cost of e-waste and carbon dioxide emissions has to be qualified and that cost must be added to the consumers of the energy, which then can drive market forces, creating a demand for energy efficient code. Even if we move to a zero-carbon energy model, we are still going to have to make significant energy efficiency improvements (UN Figures of a 4-fold increase), to deliver on our carbon commitments. The data centre providers must come up with a way to calculate energy use for each individual consumer and the Computer industry must calculate energy use for each individual application function. Fortunately the industry is aware of the issue and there are a number of organisations who are looking into how to develop more energy-efficient software.
