Institutes that store their data in the cloud instead of on their own server save a great deal of energy. A sample calculation with SURFdrive shows that the connected institutes can potentially save 60,000 kWh per hour collectively. In other words, a consolidated cloud service is 13 times as energy-efficient as hosting on individual servers.
The energy consumption in the ICT sector continues to increase year on year. Large numbers of servers are running around the clock, 365 days a year at large as well as small companies. However, ICT equipment is becoming increasingly energy-efficient. And this equipment is also being used more efficiently. A well-known method for using resources more efficiently is based on consolidation, which involves putting dispersed IT resources together. The most frequently used method for consolidation is server virtualisation. With one physical server, it is possible to virtualise several smaller servers in order to utilise the resources of the server more efficiently.
On the Global e-Sustainability Initiative website, you can see how much energy would be saved if various countries would switch from their own servers to a consolidated cloud environment. Unfortunately, the Netherlands is not on the list. Yet according to a 2014 report by Capgemini, the Netherlands is one of the leaders when it comes to the cloud, with an adoption rate of 28%. By way of comparison, the European average is 19%. Of course, these numbers will have increased somewhat by now, but the adoption percentage is still surprisingly low.
Thirteen times as economical
The more servers you put in one place, the more efficient use you make of your resources. The difference in energy consumption between a server that is idle and a server that is being used at full power is relatively small. After all, the disks continue to turn, regardless of whether they are being used or not. Accordingly, it makes much more sense to utilise a server as efficiently as possible.
SURFdrive is the cloud storage service provided by SURF for the research and education sector. At present, 54 institutes are using this service for their employees, equating to 20,000 individual users. What would the effect be on energy consumption if all these institutes were to host their storage services locally? A small calculation shows that using cloud storage services at a central location is 13 times as economical compared to a situation in which all these institutes were to host the storage services at their own site.* This calculation will be explained in more detail further on.
Power usage effectiveness
Even this figure of 13 times is still a modest estimate of the overall energy efficiency. Consider, for example, the PUE (power usage effectiveness), which represents the ratio between the total power consumption of a data centre and the power consumption of the ICT equipment. The new SURFsara data centre has an economical PUE of 1.22. The average PUE reported for data centres in the Netherlands is 1.31, according to a report from the Dutch Datacenter Association released in 2015. Experience shows, however, that the smaller the data centre is, the higher the PUE. As building management and maintenance always involves costs for cooling, lighting, and so on, a smaller data centre is always less energy-efficient. On average, a simple server room at an educational institution will have a much higher PUE.
In view of the international average PUE of 1.8, the Netherlands is a leader in this regard. However, these numbers are theoretical PUE values. The actual PUE value can be significantly higher or lower, for example due to temperature fluctuations. Field measurements show that the actual PUE is often higher than the design PUE. Even a small change in PUE can make a big difference in terms of how much more energy-efficient a consolidated environment is. This is demonstrated by the sample calculation below.
ICT equipment is becoming increasingly energy-efficient. This means that a shorter turnover period, during which a new generation of ICT equipment becomes operational, also contributes to a reduction in energy consumption. With every new generation of equipment, the various server suppliers also provide new ways of monitoring and managing power consumption more effectively. One of these methods is power management, which is also used for SURFdrive.
Power Management is a set of techniques that make it possible to save energy at server locations. The processor consumes a large part of the energy used. Thanks to dynamic frequency scaling, Power Management makes it possible to reduce the clock speed of the CPU when it is not being used to the full. For SURFdrive, this method has resulted in savings of 20%.
In this sample calculation, we assume that the institution has redundant data facilities in place. This means that all data is stored on two different physical servers. Accordingly, if one server malfunctions, there is always a backup available. This is usually the case, but not always. If there is no redundancy in place, fewer servers are needed. In the example below, we assume that the connected institution has its own sync-and-share solution with a similar level of redundancy to SURFdrive. This redundancy is in place not only for the storage, but also for the proxy servers, web servers and database servers.
The calculation is as follows. As mentioned above, 54 institutions are presently using SURFdrive, 47 of which have fewer than 1,000 users. Let us assume that each institution uses a redundant setup of 4 servers for local storage. We assume that these servers are fully used for storage and that no form of consolidation is applied**. On average, a storage server uses approximately 270 W per hour. This results in:
(4 servers * 47 institutes) * W = 50.760W
Then there are still another 7 institutions with more than 1,000 users. These institutions use more servers for local storage as well as additional servers such as database servers and proxy servers. The second proxy server acts as a failover server and consumes less power. This could then look like the following:
((4 storage * 270W) + (3 database * 200W) + (1 proxy * 200W) + (1 failover * 160W)) * 7 institutes = 14.280W
This equates to a total consumption of:
50.760W + 14.280 = 65.040W
For the present SURFdrive environment, which serves 54 institutions, the picture now looks as follows:
(5 storage * 370W + 4 storage * 270W) + (4 database *200W) + (6 web * 152W = 0,912W) + (1 proxy * 200W + 1 proxy * 160W) = 5.002W
Finally, we make use of the number that represents the ratio between the total power consumption for SURFdrive and the total power consumption that would result if all the institutions were to use their own local storage facilities:
65.040/5002 = 13,003
In other words: we can say that the use of SURFdrive is 13 times as energy-efficient as using local storage.
But we could also go one step further. Earlier, we mentioned the PUE. Let us assume an average PUE of 1.5 for other data centres and server rooms. If we multiply this number by the power consumption, the result is:
65.040 * 1,5 = 95.560
5.002 * 1,14 = 5702
95.560/5702 = 16,76
Due to the large difference between cloud storage and local storage, the difference only gets bigger.
Finally: green and grey energy. Green (sustainable) energy is generated with the help of sustainable energy sources and has less of an impact on the environment. As a result, green electricity has a small – or no – CO2 footprint. According to a report from the Dutch Datacenter Association, 79% of the energy consumed in data centres is green energy. The remaining 21% comes from fossil fuels, also referred to as grey energy. SURFdrive is powered by 100% green energy.
* This case deals with the savings that could potentially be realised based on the capacity offered to institutions by SURFdrive. For this article, no investigation was carried out to determine whether these institutions still manage servers of their own.
** In this sample calculation, servers for management purposes were excluded. If management servers were also taken into account, the result would be even more in favour of SURFdrive.
More information about SURFdrive: https://www.surf.nl/services-and-products/surfdrive/surfdrive.html
About the author
Diederik de Graaf studies System and Network Engineering at the Amsterdam University of Applied Sciences. He previously designed a visualisation platform that monitors energy consumption for the Greening The Cloud project together with a team of fellow students. He conducted the above study within the framework of his internship at SURFsara.