Background

In 1769, James Watt (1736-1819) patented his new steam engine that was 75% more energy efficient than any of the competing designs. However, he still faced a  considerable challenge convincing the hard-nosed northern English industrialists  of the economic advantages of  power generated by steam over power generated by horses.

James Watt improved steam engine efficiency by approximately 75% compared to Newcomen’s earlier steam engine design. Nevertheless, only 3% of the energy produced by coal was converted to usable energy (motion).

Watt succeeded by demonstrating that not only was he a brilliant engineer, but that he also had a marketing talent of equal measure. He expressed the power output of his engines in terms that his industrial customers using horse-drawn power could understand- Horses.  He estimated that a horse could pull 120 Pounds up a well of 220 feet high in one minute, and subsequently an engine with this capability had the power output of One Horsepower.

Using horsepower it became relatively easy to compare the costs of producing power by a 10  horsepower steam engine  to that by ten horses. It became clear that the capital and operating costs of steam engines, coal, housing and maintenance, were far less than the stabling and  feeding costs of horses. Inevitably, Watt won the debate.

Under the  EU Energy Efficiency Directive (EED) Recast and similar initiatives around the world, data centres are under the shadow of greater scrutiny and tighter regulation in an effort to reduce their energy consumption and environmental impacts. One major consequence  is that centres in the EU will have to measure and show how productive they are in their energy usage in a mandatory annual audit.  This level of transparency is essential if the EU is to meet its target of a 11.7% reduction in energy consumption by 2030 relative to the forecast energy consumption for 2030 made in 2020.

Productivity has many aspects but to illustrate how unproductive data centres have been in the past, the 2015 Stanford Koomey report found 30% of servers in a survey of data centres were inactive (comatosed) for 6 months or more, having not provided any services or computing in this period.  If this was representative of centres in general, it would represent a waste of $30 Billion in I.T investment worldwide. Furthermore, as servers typically consume half of their maximum power consumption while idle, it highlights the potential for enormous energy savings, and scope for minimising environmental impacts at the discretion of the industry. Globally, it is estimated that centres account for 3% (1,500 million tonnes approx.) of global CO2e emissions.

More recent reports don’t make encouraging reading. The Uptime institute, an independent global digital infrastructure authority, in their 2024 review of data centre practices worldwide , reported that 60% of centres still don’t track server utilisation.

Assessing productivity or efficiency, would be quite straightforward if there was some standard unit of computing. Then, like Watt introducing Horsepower, it would be possible to compare the energy cost of producing one unit of computing. Unfortunately, with the huge variety of computer uses, services and applications, coupled with the complexity of computer systems and configurations, it is impossible with so many variables to define a standard “unit of computing”.

Different Horses for Different Courses

Nevertheless, a solution does exist if we accept that there are “Different Horses for Different Courses”, and permit users and providers to define metrics that are specific and relevant to their own applications or services. Being User-defined, overly erudite or technical metrics can be confusing but this can be circumvented by more using comprehensible and meaningful performance metrics specific to the application. For instance, for an on-line shop company, the metrics could be the energy consumed and carbon footprint per 1000 transactions ordered. For a website video-streaming business, an appropriate metric would be the energy and carbon cost for processing and transmitting 100,000 frames, while for an insurance company there may be similar metrics but for 100,000 data base queries.

While there are some accepted international standards for data centre operations – for example, Power Usage Effectiveness (PUE) – there are few other measures of data centre processing performance that are globally accepted.

In simple terms, PUE is the ratio of the total energy consumed by a centre relative to the that consumed by the I.T equipment. So, a PUE =1.5 indicates a 50% energy overhead (cooling, air-conditioning etc) to every kWh consumed by I.T (servers, memory, routers etc). PUE has become almost a status symbol for the industry with centres boasting of  their small PUE. Still, this simple analysis can be deceptive. A centre with a perfect PUE=1 indicating all energy is only being consumed by its servers could be wasting all this energy on servers doing nothing.In fact, some critics of the metric have proposed an alternative acronym for PUE- Pretty Useless Entity.

There are a multitude of advantages in employing user-defined metrics in data centre energy management, among them are:

1. The actual financial, energy and environmental cost of services can be accurately measured and understood by all organisation management levels.
2. System upgrades can be targeted for maximum ROI.
3. The cost benefits of upgrades can be quantified.
4. Customers and clients availing of services can be informed of their Scope 3 emissions.
5. Use of metrics in the annual performance audit is convincing evidence of an organisation’s commitment to a sustainability agenda.
6. The financial and environmental cost of performing the same tasks on different computing platforms, with different tariffs and climatic conditions can be readily assessed.

A Practical Solution to the Problem

Core to a user-define metric solution is a mechanism that measures in real time the energy consumed by the individual computing tasks and processes associated with the particular metric. This can not be done be simply sticking a meter on the power cable of a server. Most services and applications share servers (virtualisation) and memory resources with several other independent processes. Therefore, the energy measuring mechanism needs to track the amount of each resource used by each individual application or service.  In the case, of the insurance company data base query example, this would involve tracing the execution of the data base software on all its computing platforms, its memory activity, and the relative amount of each resource utilised.

In the School of Computer Science, University College Dublin we have found a solution through technology researched, developed and commercialised by one of its spin-out campus companies, Beeyon. Using Beeyon’s data centre energy performance technology Papillon, all servers, applications and services are monitored in terms of energy consumption, carbon emissions and efficiency. Typical of many educational establishments and large organisations many servers are distributed around the campus, but as a software-based solution requiring no physical devices or retrofitting, Papillon can be installed in minutes and operate securely behind the user’s firewall.

As an integral component in the School’s objective to be carbon-neutral by 2027, Papillon is project to reduce I.T related emissions by 30-40% and operating costs by a similar amount.

For more information on Papillon and the School of Computer Science’s Carbon-neutral strategy contact damian.dalton@ucd.ie