Why the devil’s still in the detail in the race for energy efficiency
With UK governments since 2010 not always prioritising energy efficiency, even reducing funding measures with full support from high-profile members of parliament, one might wonder whether the criticality of energy efficiency has been overemphasised.
Energy efficiency is crucial for driving, among other things, sustainability goals, as most cloud and datacentre services providers readily agree. However, as Tim Loake, vice-president of the infrastructure solutions group at Dell Technologies UK, says: “It [energy efficiency] is a very nuanced topic – and as soon as you think about it through different lenses, it gets more complicated.”
One issue may be that actually achieving greater energy efficiency means a move further beyond a simplistic focus on “energy efficiency” ratings and labels – for instance, on hardware.
Loake says the “easy answer” is typically that energy should equal compute load out – or “as close as you can get” – and means aligning with artificial intelligence (AI)-enablement, cooling advances including liquid cooling, and future tech innovations.
Traditionally, that sort of progress has focused on datacentre Power Usage Effectiveness (PUE) score, on which progress now appears slow.
Meanwhile, not only has power become more expensive, but total energy consumption, with the concomitant greenhouse gas emissions, depends partly on the sources of that energy, as well as all the energy that’s gone into related products and services over their entire life cycle.
“And we’ve seen, because of the spiralling energy prices, that now you’re having the time window shrink by which it makes sense to actually replace your equipment,” adds Loake.
Previous replacement timeframes of three to five years or so no longer fit the bill. Assets become uneconomical earlier – and datacentres or providers may or may not have been able to negotiate with suppliers.
Loake has seen customers replacing kit that’s just two years old because it will slash their power bill by 40% or more. “That precipitates new issues,” he points out. “How long am I dual-running while I move my workloads from one place to another, increasing my power use while I do that.”
Then there’s rehabbing, rehoming or recycling the old kit responsibly, and the overall cost of the increased amount of kit. How much plastic is in the units, and how much embodied carbon across the whole lifecycle, and how much can be recycled or reused in a year or two? “It’s not straightforward,” he says. “Overall, I think most organisations are still driven by their profit-and-loss sheets.”
That said, Cambridge University as an HPC customer has built out a new GPU cluster, and Dell helped them turn down GPU performance by 9%: reducing power consumption by a huge 40%.
That machine is now number three in the “top 500” most energy-efficient supercomputers, says Loake.
“It’s working and thinking carefully about what you do that allows you to make these changes,” he adds. “And with our server technology, we’re always thinking about how we can save energy and also make the equipment more suitable for recycling. For example, we took the hard drive caddies and took all the plastic off of them because the plastic just made them look nice. It served no functional purpose.”
All-metal caddies can be easily melted down and remade. Equally, systems management platform sensors pinpoint hotspots and dynamically adjust the fans independently. It then becomes about how you can save energy without necessarily incurring more cost and without increasing your carbon footprint due to existing estates, says Loake.
We need better ways to compare offerings
Right now, too, many systems are labelled based on peak consumption, which might not reflect actual operating conditions, for instance. Offerings are being evolved among the Tier One suppliers for this. Dell is working on labelling the systems in better ways, which will enable customers to more easily compare apples with apples, and oranges with oranges.
“Because it’s dependent on what workload is running, how much power it needs,” says Loake. “We talk about CPU consumption as 250W a socket at peak utilisation, but very often it shouldn’t be using that unless it’s an HPC system. So, you have to think about what you are using today.”
Gisli Kr. Katrínarson (Gisli KR), chief commercial officer (CCO) of Nordic datacentre services provider atNorth, says it’s “hugely important” to differentiate between power efficiency and energy efficiency concepts. It’s also requisite to apply related thinking across entire systems, going beyond the datacentre, he says.
“Power efficiency is something we’re used to seeing broadly used by enterprises to meter what datacentre they should choose to provide them with the lowest TCO,” says Gisli. “How much electricity, now becoming more of a scarce resource, is going towards compute versus cooling? Are you reusing heat, or otherwise mitigating energy that the datacentre uses or is a waste in the datacentre?”
One atNorth datacentre in Stockholm is expected to potentially heat perhaps 20,000 nearby homes – an example of how a different understanding of energy efficiency might, taken as a whole, deliver energy and emissions savings.
Often such possibilities might not be looked at closely enough or fully understood – especially when bearing the cost of building any related infrastructure might mean a temporary competitive disadvantage.
What’s needed is industry collaboration to truly “level up” with energy efficiency, making better use of resources. “In Sweden, we are working with utility provider Exergi, where they are basically enabling companies to mitigate heat to the central heating system,” says Gisli, adding that a regional resource park has been mapping out circular economy activities.
By better understanding who is using what, productive areas for future collaboration may be teased out.
“In all respects, it makes more sense, for example, for heavy computing infrastructure to be moved closer to energy sources. Because when you move the energy to the region, you get heavy, wasteful losses of energy,” he says. “Then people get a certificate and brag about it being green – it’s utter nonsense.”
Criteria are becoming much more stringent, and organisations increasingly understand the value of applied sustainability metrics in procurement processes for better TCO, instead of looking at a here-and-now assessment of energy use that doesn’t take into account the whole lifecycle of IT equipment, data and software.
It’s time to rethink all that, he says, and get buy-in for better services at other levels of the business, such as the finance department.
“People (including customers) are now demanding answers to questions that are quite challenging. What is the lifecycle for not only my IT equipment but my data? What does my workload really need?”
Tim Allen, head of engineering, devops and cloud at web and mobile product developer xDesign, points out that some feel it’s better to spend time making one’s computers faster, increasing network speeds and storage capacity, to do more with what you already have.
“But that view fundamentally misses the point, whatever your opinions on climate change and net-zero,” says Allen. “Using what we have more efficiently is a key tenet of developing lean systems.”
The importance of efficiency
Most users do not require ever faster processors or networks – storage is already effectively infinite. Yet clients continue to want an efficient service with a cheaper operating cost.
Traditionally companies make profit by providing new and innovative services and by doing so as efficiently as possible. “However, on the software side, as time has gone on we’ve got ever faster computers, more memory, faster networks, and as software engineers, we have become lazy,” says Allen.
Considerable energy efficiency gains could be made by coding better from the start. Yet that’s going to mean better training and education of programmers and related professionals. This is not only to deliver the required technical skills – including less reliance on tools and compilers, but to communicate why lean code is important.
“In the late 1990s, one of my lecturers said you only need to learn this vaguely efficient sorting mechanism because computers are getting faster, performance is getting better,” he adds. “And now we’re in the big data era, and that’s patently not true.”
Allen says he’d love to see the efficiency metrics of “certain big systems” such as Amazon or Netflix, and that it would be interesting to experiment with slightly slower responses for more environmental benefits.
Would people be happy with that? If such an approach could be sold to the end user, by speeding up software you could potentially deliver even more energy efficiency, says Allen.
The “big disincentive”, he adds, is always likely to be cost. But why can’t more providers try to give something away for free – to help drive more energy efficiency? Either way, systemic change is needed in the way we think about resource efficiency and how to achieve it. “We need the consensus of enough experts,” says Allen.
Cause for optimism ahead
Josep Alvarez-Perez, partner and head of the xTech division of global consultancy Bip UK, is optimistic about the industry’s ability to ultimately get a grip on the diverse threads of energy efficiency, however – pointing to the evolution of tech, from chipsets and up to the renewables transition of the hyperscalers.
“And in the past two or three years, they have adjusted the fees to incentivise more efficient use of resources,” he says.
Beyond evangelising about the potential for cloud and better application design in assisting optimum slice-and-dice of resources for greater efficiency, Alvarez-Perez agrees that the diversity of certifications making sustainability or energy-efficiency claims can make choices confusing for customer organisations.
“Yes, but we are in the early days of environmental sustainability and governance,” he says. “At least we are now having some options – this is, I think, a good trend.”
Alvarez-Perez says more progress can be expected, although perhaps not with trenchant “geopolitical issues”, as the Covid pandemic fades. Companies are increasingly investing, and sustainability will increasingly come to the fore.