Dividing lines: EU bid to curb server energy use has the European datacentre community split

Proposed guidelines

The Enterprise servers and storage portion of the directive is covered off in Lot 9, with the EU proposing to set guidelines on how much energy products within this category definition consume when operating in an idle state.

According to EU estimates, the implementation of Lot 9 could collectively result in annual energy savings of approximately 9TWh by 2030, with 2.4TWh of these savings attributable to curbing the amount of power used by idle servers.

To put that 9TWh figure into perspective, the draft of the regulation claims this is on a par with how much energy Estonia uses over the course of a year, based on 2014 figures.

Differences of opinion

The prevailing view among stakeholders is that any effort to curb the continent’s energy use on such a large scale are welcome, but it’s the EU’s proposed use of the idle energy metric to achieve these savings that has proven to be so contentious.

A four-week consultation on the proposals back in July 2018 saw IBM, Dell-EMC and HPE all query the rationale for using the metric, claiming that idle energy measurements are an ineffective means of determining how energy efficient a server truly is.

In fact, Kurt Van der Herten, EU environmental policy program manager at IBM, says – in a statement to Computer Weekly – that the directive’s proposed methodology could end up driving up the energy use of datacentres, rather than reducing it.

“There are elements that may have the consequence of decreasing the energy efficiency of and reducing the power consumption savings from datacentres contrary to the intent of the Eco-Design directive,” he says.

“The proposal to set a limit of idle power consumption of servers will result in the deployment of a larger number of less efficient servers, higher energy use, and poorer datacentre energy performance.”

Server use neglected

It is further claimed, in a separate consultation response by HPE, the Directive neglects to take into account how servers are used within datacentre environments, given its focus on measuring the idle energy use of each individual appliance.

“The current focus on idle addresses the individual product one, and fails to recognise how servers are used to manage multiple workloads and utilisation levels and how it can be done in the most efficient way,” says Pieter Paul Laenen, compliance manage for Europe, Middle East and Africa (EMEA) at HPE, in its written response to the four-week consultation.

 “In essence, [by] settling idle limits for individual servers which are too tight for new high performance servers,  [it is our view] that this will result in EU datacentre operators being forced to use more low performance servers at a higher total energy use.”

Actively opposed

This sentiment is broadly shared by all three suppliers, who have all separately made a case in their consultation submissions for the idle energy metric to be dumped in favour of an alternative active efficiency measure. Because, in their view, it provides a better overall picture of how well these appliances perform.

“Active efficiency remains the optimal tool to remove the least efficient servers, driving energy efficiency not only in enterprise datacentres but in small closet installations as well,” says HPE, in its submission.

This is because it not only takes into account the amount of energy used when servers are running idle, but also how much power they consume when in active use too, says HPE.  

The suppliers’ claims have won the support of TechUK, who further asserts that some of the most efficient servers on the market consume relatively higher amounts of energy when idle, but that does not mean they should be precluded from sale within the EU. 

Susanne Baker, head of programme, environment and compliance, at TechUK says: “Servers have become better performing and are more efficient when operational, the trade-off is a slight increase in idle energy.

“Overall though it results in energy reductions. Measuring server efficiency by only using idle power metrics will see the most efficient and best performing servers banned from the EU market,” she says.

This is based on the theory that, once the Directive comes into force, datacentre operators will opt for servers based on how much energy they consume at rest rather than how much they use when performing a given task – and, in turn, this could result in servers being deployed in datacentres that consume more energy overall, and are, for that reason, considered to be less efficient.

An industry divided

This assertion, put forward by those opposing the use of the idle energy metric, is roundly contested by a number of datacentre industry stakeholders, including academic researchers and analysts, as well as some other members of the server supplier community too.

Some have private expressed misgivings to Computer Weekly over the motivations behind Dell-EMC, HPE and IBM’s decision to publicly condemn the EU’s use of the idle energy metric, with some suggesting their dissatisfaction is borne out of a need to meet their own commercial interests.

After all, in its consultation submission, Dell-EMC claims the EU’s mooted energy saving projection figures would come at the expense of product availability, as around 76% of the servers currently on sale would have to be phased out for failing to meet the Directive’s “aggressive idle power limits”.   

All three of the manufacturers in question retain a sizeable hold on the EMEA server market at present, but have also seen their dominance challenged in recent years by the hyperscale cloud community’s growing appetite for datacentre kit made by white-label, original design manufacturers (ODM).

It is also worth noting there are manufacturers out there making high-performing, energy efficient servers that are widely used in datacentres across Europe, who have not seen fit to respond to the consultation at all, because their kit falls well within the idle energy limits.

“There is a huge rise in a new generation of servers that are based on open standards, such as the Open Compute Project ones, that already well exceed anything the legislation is requiring,” says Rabih Bashroush, a reader in distributed systems and software engineering at the University of East London’s (UEL) School of Architecture, Computing and Engineering, on this point.

As part of his academic work, Bashroush recently completed the 36-month, EU-backed EURECA research project, which focused on helping public sector datacentre operators pinpoint areas where cost and operational efficiencies can be made within their facilities.

He is very much in favour of what the Directive is trying to achieve, as well as the EU’s decision to clampdown on the amount of energy servers consume when running in an idle state.

“Low server utilisation is perhaps one of the key problems we have when it comes to energy waste [in datacentres]. When servers are running idle (i.e. doing no useful work) they still consume anything between 30%-to-70% of their energy, which also requires the underlying datacentre power and cooling infrastructure to be operational (and consuming energy),” he says.

“According to the research findings from our EURECA work, the average server utilisation in Europe ranges between 15%-to-25%, with the occasional high performer averaging 30% or so.

“To help reduce energy waste, we ought to do something about idle state energy consumption, and that is what the EcoDesign legislation is trying to do,” he adds.

And while the Directive is focused on curbing the datacentre industry’s overall energy consumption using idle energy caps, it is unlikely to be the EU’s sole aim, says John Goodacre, a professor of computer architectures at the University of Manchester, and founder of UK-based converged infrastructure startup, Kaleao.

“I suspect the whole point of this legislation was to drive innovation and persuade these guys to do things differently. There is not a market pressure that motivates them [at present],” he says.

“It is a historical design choice in a sense that if you have to be low power, you put in those features, and if you don’t, you haven’t, and the Directive could promote innovation and change over time.”

Encouraging innovation

This is a view shared by John Laban, a European representative for the Open Compute Project (OCP), a Facebook-backed industry initiative whereby participants share datacentre and server design concepts with each other and the wider IT community to encourage innovation.

Instead of focusing on how onerous the idle energy limits are, manufacturers should be using the Directive to rip up the server design rulebook and revamp their own product roadmaps.

“Whether we like it or not, it is very clear that Europe needs to do something to lower the energy usage of its fast-growing datacentre industry. The products and technologies to do that are already available as open source hardware designs,” he says.

“So I actually think that the EU is giving us a great opportunity here to start using innovative datacentre hardware that will make it possible for a typical datacentre to reduce the energy consumption of idle servers by at least 50%.”

The final countdown

At the time of writing, the Directive was in the final stages of being formally adopted by the European Commission and enforced, after it was cleared with a majority vote by the Regulatory Committee to proceed on 17 September 2018.

Its contents will now be subjected to three months of scrutiny by members of the European Parliament and Council, and – while they cannot amend its wording – the draft can still be opposed.

For supporters of the EU’s preference for using the idle energy metric, the outcome of the vote is being treated as a significant win, particularly as past revisions to the Directive have led to accusations that its content has been significantly “watered down” over the course of its successive drafts. 

“The legislations excludes all HPC servers, servers with integrated APAs and high resilience servers, plush many others (based on the number of cores running the same operating system and number of ports, for example),” says UEL’s Bashroush.

“If anything, the legislation has been watered down so much already by the Commission due to pressure from certain industry players, diluting it any further by removing the idle power limits will defeat the purpose of the legislation and will mean a major opportunity is missed to reduce the energy waste in datacentres.”

Late opposition a mistake

Furthermore, Laban’s colleague and fellow OCP European representative, Robbert Hoeffnagel, says any subsequent moves to oppose the Directive at this late-stage would be a mistake.

Particularly as, in his opinion, it could have as big an impact on the European datacentre industry as the introduction of the Power Usage Effectiveness (PUE) metric did more than a decade ago.

In that time, the metric has gone from being a concept pioneered by The Green Grid in 2007 to becoming a metric that is widely used across the industry by operators to benchmark the energy efficiency of their facilities, and pinpoint areas for continued improvement.

“Look at what happened with PUE. It was a controversial metric and in the beginning hardly anybody really knew how to do the calculations, but it grew into something quite powerful when datacentres started to recognised they could use PUE for marketing purposes,” he says.

“That started to drive investments – little by little – in reducing overall energy usage by getting rid of the low-hanging fruit. Maybe in time we will see the same trend here.”

Conciliatory role

And, with the Directive getting closer to coming into force unopposed with each passing day, TechUK’s Fryer says she sees its role in the debate changing, and becoming more conciliatory in nature, to ease the tensions between those for and against the proposals.

“Over the last few weeks, we have been sounding out the wider sector on this issue to try and understand the gap between industry and the Commission, [as] some academics take the view that we can have our cake and eat it on the idle power front. This view is not shared by the major server manufacturers,” she says.

“Much seems to depend on how people anticipate how the sector and its business models will evolve:  the prevailing view has been that we will see greater consolidation, continuing the existing trend towards larger, more powerful machines.”

“Others envisage a scenario where the trend is the opposite – towards smaller, lower power devices within a distributed or Edge infrastructure.  

“I suspect the data centre landscape of the future will accommodate both models – in which case it is even more important that regulation is appropriately targeted. Time will tell,” she concludes.