Many organisations are investing in making their Internet and IT infrastructure more energy efficient. But in the search for sustainability, a hardware and data centre-centric approach may be missing the point.
Recently, Adaptive Capability was asked to comment on an Australian government initiative to improve the energy efficiency of data centres. The discussion report proposes a number of sensible suggestions including applying energy efficiency ratings and/or minimum standards to data centre facilities and the IT equipment they house. However, we wondered if these approaches obscured a much broader opportunity for transforming the IT industry to a more sustainable footing.
What we’re seeing in the data centre space is a run away freight train of more and more processing power and storage globally, supporting all the amazing web apps that people suddenly cannot live without, coupled with corporates collecting vast amounts of “big data” to try to get new insights about their customers and products.
Fundamentally, energy demand in data centres is a behavioural problem linked to our use of technology. While we’re not going to be able to change that very easily, buying more energy efficient servers or improving the Power Usage Effectiveness (PUE) of our data centres seems to to be like putting a band aid on a cancer victim. We think, however, that without trying to tackle the apps/big data juggernaut, there is something that governments could do that might have a more sustained impact
As a long term representative of a technical working group for the Australian government’s NABERS (National Australian Built Environment Rating System) we note that the data centre rating tool released in 2013 made significant compromises in attempting to come up with a measure of the efficiency of “useful computing output”. Measuring hardware efficiency improvements in metrics such as megaflops or disk I/O (input/output) per Watt (W) is relatively easy but the working group didn’t manage to figure out a way to build a viable, assessable metric that would serve as a proxy for, for example, “emails delivered per W” or, more usefully, “tax returns processed per W” or “Facebook posts per W”.
That’s where we think a lot more work is required: tackling the efficiency of computer software itself (as well as the way people use it).
The inefficiency of software is, we believe, directly linked to the continued realisation of Moore’s Law, which has effectively led to a doubling of compute power (typically for about the same or lower cost) every couple of years for the last 50 years. This has created a software development culture that encourages “bloat-ware”: there is no need for coders to cut resource efficient code because a faster computer or device with more RAM (Random Access Memory) and storage is released every few months. Meanwhile there’s money to be made in adding new features (whether they’re needed or not) and churning out new versions of ever more resource hungry software every couple of years.
Inefficient code (and lazy operating systems that allow a build up of “system detritus” and memory leakage) benefits the hardware suppliers since it creates an impetus for people to upgrade their devices on a regular basis, cementing a mutually-beneficial relationship between the hardware and software worlds.
Many organisations typically figure on replacing compute equipment within three years (and many people replace their mobile devices more frequently in line with two year phone plans), creating a mountain of eWaste with huge ecological impacts in terms of embodied energy, CO2e emissions and resource depletion. Whereas you’ll typically get at least 50 years of economic value out of a building, meaning the embodied energy is usually significantly lower than the energy in use, in the case of computing hardware the equation is probably a lot closer to parity or worse and the lifecycle operating energy costs of a server typically exceed the purchase price.
If we are serious about tackling ICT energy efficiency we need to start with the software industry. Universities should be teaching resource-efficient software development. There should be measures to encourage applications and operating systems that run comfortably on older hardware. Major software and hardware companies (which are typically US based, so international cooperation would be required) could be investigated to see whether there is any evidence of anti-consumer collusion in perpetuating the Moore’s Law-driven spiral of software upgrade necessitating hardware upgrade. Leaders in sustainable computing (i.e. maximising the longevity and efficient use of compute resources) should be identified and celebrated.
Consumer education is also important to help people understand how their use of technology is leading to inefficiency. Something analogous to the former Australian Labour Government’s “black balloons” campaign but associated with the energy costs of each photo they upload to Facebook; each Google search; even the extra bytes of storage associated with their email signature file and the near ubiquitous “Please consider the environment before printing this message” sign off.
We recommend the objective of government policy in this area should be, on the one hand, to minimise the amount of hardware and associated infrastructure required to perform a particular function, but also, critically, to prolong the economic life of that investment in hardware and infrastructure.
Talk to Adaptive Capability today about ways to safe guard your organisation’s future.