Very often cloud processing simply shifts energy usage from your own data centre to someone else’s – often in another country. While this might make you feel better, it does nothing for overall planetary energy efficiency or carbon emissions.
This week British data centre energy consultant John Booth was in Australia. At a talk in Sydney for the Australian Computer Society he gave a number of reasons why cloud processing is not necessarily any more energy efficient than local processing:
- Greenness is not just about energy use. It includes the embodied energy in the manufacture of a device. Cloud does nothing for this.
- Not everything in an organisation can be moved to the cloud – some processing must always occur locally. Energy efficiency will then decrease, because less work is being done by the same equipment.
- An enormous amount of energy is used for the network capacity that enables cloud computing. This is never accounted for, but studies have shown that Australia’s telecommunications infrastructure accounts for more than 10 percent of all energy used on ICT in Australia. This energy usage is “hidden in the utility.”
- Cloud computing also means increased usage of local data communications devices like switches and routers, all of which consume energy.
- Even if cloud computing is more efficient, there is no way of measuring its efficiency or quantifying the gains. No-one has properly measured pre and post cloud, so nobody really knows if there is a reduction, an increase or a status quo.
To Booth, the bottom line is a quote from well-known analyst Andy Lawrence, research director of the 451 Group. “Cloud doesn't save power but displaces it,” says Lawrence. “Ultimately, roughly the same power is drawn from the grid, just by different companies. So it's no greener. Cloud is more about dealing with company-specific issues than planetary ones.”
None of which is to say cloud processing may not be more efficient. If your cloud processing is all taking place in a super-duper energy-efficient data centre located in a cool climate, running off hydroelectricity and with lots of free cooling, it will obviously be energy efficient. Even when you take into account the overheads of data transmission, it is going to be more energy efficient than processing performed locally in an old data centre run on dirty power.
Booth made the very valid point that virtually all Australian data centres take their energy from the electricity grid (cogeneration and trigeneration are in their infancy). That energy is largely generated from coal-fired power stations, meaning it is very carbon-intensive. So any Australian-based cloud is never going to be energy efficient.
Over the last few years I’ve been involved in a few exercises that have attempted to quantify energy usage of corporate computing. It’s a very difficult area. At the same rate equipment is becoming more efficient, we are using so much more of it that any gains are being more than cancelled out.
It energy efficiency is a vast problem, which I shall return to in future columns. John Booth’s contribution is significant – cloud computing might help – but it might not, either. Whatever the case, it is only a small part of the answer.