Verne Global

Data Center | Insights | Sustainability |

30 January 2018

HPC research vs Gangnam style - Measuring the ‘social worth’ of data centers

Written by Andrew Donoghue (Guest)

Andrew Donoghue is a technology writer specialising in data centers and critical infrastructure. He has worked for analyst companies such as 451 Research and has also held senior editorial roles at various tech publishing companies. You can follow Andrew at: @andrew_donoghue

It might be time to seriously question, and measure, if the energy and resources used by data centers are of equal value to society.

The dramatic fluctuations in the price of Bitcoin recently won’t have escaped many people’s attention. Some speculators and investors will have of course benefited, but critics have also cited the series of booms and busts as clear evidence of the inherent volatility of Bitcoin and other crypto-currencies.

Creating each new Bitcoin requires increasingly large amounts of compute power and energy (not all of it in traditional data centers of course, and not all of it based on renewable grids as in Iceland). Calculating exactly how much compute capacity is challenging however as the complexity of the process to create a unit of Bitcoin is constantly shifting.

According to the Digiconomist Bitcoin Energy Consumption Index, the annual energy use of the entire Bitcoin network – in so-called ‘hashing centres’ – has nearly doubled in the last three months to approximately 42 TWh. Put another way, that’s close to the energy use of New Zealand or 0.2% of the world’s total electricity consumption. If those figures are accurate, Bitcoin is significantly more energy intensive than an established payment network such as Visa while handling a fraction of the transactions.

But despite the growing energy demands of cryptocurrencies, it’s probably unfair to single them out when it comes to questions over social worth. Arguably huge swathes of online activity, and as result data center capacity, are not adding much to the common good or moving our collective ‘brain trust’ forward.

Online video is an obvious target. UK data center expert Professor Ian Bitterlin has written and presented convincingly on this subject. A few years back – when the Gangnam Style video was at the height of its viral madness – Bitterlin dug into the energy consumption of all those YouTube plays. The results were startling to say the least. On average 0.01kWh of energy is consumed in carrying 1MB over the Internet while the energy consumption of consumer devices is around 0.001kWh for one minute of video streaming, according to estimates cited by Bitterlin.

The Gangnam Style video was played approximately 1.7 billion times in its first year (it’s now at 3 billion plus). Based on a file size of 17MB, Bitterlin estimated that this equated to around 298GWh, or 35MW of constant power consumption, in one year. Or to put it another way, roughly the same energy use as the annual electricity consumption of the African Republic of Burundi, which has a population of around nine million.

But inane content isn’t obviously the only potential waste of data center kWhs. Even if you just count the legal online activities of questionable social worth - such as a lot of social networking, pornography and gambling - that’s an enormous data center and energy footprint. Criminal activities – including those conducted on the Dark Web – are even harder to quantify but probably have a sizeable energy and carbon footprint also.

The fall-out from the 2008 financial crash also demonstrated that an unhealthy proportion of global financial services – a good proportion of it enabled by advanced IT - was little more than speculative gambling that actively damaged societies.

But if all of these online and IT-driven activities are socially questionable at best then what sort of data center activities are more justifiable? A good example might be research – using High Performance Computing (HPC) - in areas such as healthcare, energy, and space exploration. These energy intensive workloads, and the facilities that support them, should arguably be categorised and scrutinised differently to other forms of IT. In some respects they are; they are often the recipients of public sector funding but then so is IT-driven military and weapons research.

So what’s the answer? Making arbitrary decisions about the social worth of online content and other IT-driven activities is obviously highly subjective and also a very slippery slope towards extreme censorship and worse. However, it’s an interesting thought experiment to at least question whether there shouldn’t be at least some attempt to quantify and qualify the intrinsic worth of a workload, or an entire data center, in relation to its energy and carbon intensity.

The data center industry has largely united around the Power Usage Effectiveness (PUE) metric as a way to measure how efficiently facilities turn energy into compute services (and heat). The efficiency and scale of most large hyperscales are beginning to render, what was an often gamed and manipulated metric to begin with, increasingly meaningless with PUEs down at 1.02 or less.

The hunt has been on for several years for a worthy successor to PUE that looks not just at the data center IT versus facilities overhead but at the actual IT services or ‘useful work’ produced by the facility in relation to overall energy use.

The Green Grid and others have floated numerous ideas but none have really stuck. The closest we have probably come is the Digital Service Efficiency (DSE) dashboard developed by eBay. DSE tried to present a ‘miles per gallon’ equivalent for data center services but ended up being more of a smartly presented collection of measurements than a single metric. And outside of its use in eBay – perhaps mostly for marketing purposes – it failed to gain much traction.

Given that background, developing a social worth metric would be a challenge to say the least. Some established models do exist for assessing social worth more generally - such as the ‘social earnings ratio (S/E)’ and ‘social return on investment’ (SROI) - but there is not much precedence for these metrics being applied to IT services or data centers.

It could also be argued that the type of energy being used in data centers with questionable social worth should also be factored in. Not much of YouTube’s content could arguably be defined as educational per se but Google has invested billions into renewable energy projects to directly power or offset the energy used by its data centers. If the energy is additional, or renewable perhaps it doesn’t matter so much that is being frittered away on cat videos?

Ultimately, applying a social worth measure to data centers is unlikely to ever gain much traction; unless that is energy costs and availability become a significant issue in the future. A punitive universal carbon tax might also have an impact but that again looks unlikely for the foreseeable.

However, even if a social worth metric never emerges, campaigners such as Greenpeace, as well as government regulators, will continue to scrutinise the activities of large data center operators. Perhaps one day that may include asking whether at least some of those data centers out there might be put to better use.


Sign up for the Verne Global newsletter

Opinion, thought leadership and news delivered directly to your inbox once a month.