Data center energy consumption, Taiwan, climate change
Server racks in server room cloud data center. Datacenter hardware cluster. Backup, hosting, mainframe, mining, farm and computer rack with storage information. 3D rendering. 3D illustration

A recent article by Wired discussed carbon footprint emissions upon streaming Netflix using a measuring tool called DIMPACT, which indicated that an hour of film streaming releases a little less than 100 grams of carbon, encompassing the process for data to reach us from the data centers.

Certainly, these data-intensive digital services require enough space for data to be safely stored. Normally, they are hosted on facilities provided by cloud computing services, such as Amazon Web Services, Microsoft Azure, and Google Cloud that store them in data centers. AWS, for example, hosts a vast amount of Netflix data.

This raises the question of how impactful these massive energy-guzzling data centers are on various levels. Why have they become so important to businesses and how much energy do they consume? 

Why do we need large data centers?

As the world gets ever more connected and digitalized, data-intensive technologies such as artificial intelligence and the Internet of Things, are all centered around computer systems. By outsourcing IT infrastructures to professional data centers for data storage and computing, businesses are less prone to local power outages causing a direct productivity loss. 

These large data centers provide cooling, connectivity, and security, services that can often be too costly for organizations to realize on their own.

But are data centers that bad?

In 2010, the total energy consumed by global data centers is equivalent to roughly 1% of all electricity used worldwide.

That data was cited from research published on Science last year. Further estimation showed that in the following eight years leading up to 2018, that proportion remained roughly the same despite the data infrastructures had grown remarkably.

In eight years, the amount of computing done at data centers increased sixfold (550%), but the energy consumption grew by a mere 6%, and server energy usage increased only 25% worldwide (owing to growing virtualization which doesn’t consume server energy), an unproportionate increase of energy consumption was observed, thanks to processing efficiency improvements. 

Data shows energy efficiency is improved

The energy consumption of installed data storage per volume had dropped, as well as slower expansion in the number of servers required to hold computing instances due to growing virtualization. 

Another key index worth tracking is the power usage effectiveness (PUE), an index used for measuring the effectiveness of data centers. The closer PUE value approaches to 1, indicates the energy is used more efficiently at data centers where more energy is allocated to computing resources, instead of maintaining the IT infrastructures, such as the cooling systems.

According to a survey conducted by Uptime Institute, the global average had improved from 2.5 in 2007 to 1.59 in 2018 and has since remained the same.

Are centralized data centers better than small traditional data centers?

A short answer is, yes.

Big tech companies are also the largest buyers for renewable energy and have the capability to optimize computing infrastructures’ efficiency. Google’s data center PUE has dropped well below the global average reaching 1.1, and Microsoft had experimented in building their data center in the sea to minimize cooling system energy consumption.

Data center facility providers also tend to set carbon emission goals to attract business partners and are often held accountable for their business gains. According to a report by Wired, three world’s largest data center providers, including Amazon, Microsoft and Google have all set goals to go carbon neutral. Microsoft announced in a blog last year to be carbon negative by 2030.

As a member of RE100, Google said in a blog that they have purchased enough renewable energy to support electricity usage throughout the year for three consecutive years since 2017 and are working towards complete carbon-neutralized data centers.

Conventional data centers, which typically have a PUE of 2, accounted for 79% of compute instances in 2010, are shifting towards cloud services and hyperscale data centers which now carry 89% of compute instances and are more efficiently operated in terms of cutting-edge cooling-system and power supply, holding roughly a PUE of 1.2.  These hyperscale data centers are super-efficient and organized for information storage.

Data centers in Taiwan

Renewable energy currently accounts for 5.8% of the power systems in Taiwan. The Taiwanese government set the goal for green energy to reach 20% by 2025 and shut down nuclear energy plants.

Google had bought an output of 10MW power to be produced from 40,000 solar panels at a fish farm in Tainan, southern Taiwan, being the corporate’s first power purchase in Asia. One of Google’s data centers in Changua County, middle of Taiwan, had a PUE of 1.14 in the second quarter of 2020.

After Google built three data centers in Taiwan, Microsoft had also announced to establish its first cloud data center in the country last year being its largest investment in Taiwan in recent years.

It is worth considering, however, what impact will be brought by data centers’ electricity usage to the city’s power grid and how green energy must be adopted to prevent greater impacts.

In the future

The ICT industry is undoubtedly going to continue growing in the 2020s with data centers using up a larger portion of the industry’s electricity, while that is the case, ICT is also the dominant industry with the most corporates that commit to buying renewable energy.

Reducing energy consumption really depends on the physical limitations. Moving old traditional data centers to hyperscale data centers decreases energy usage. Other measures include enhancing power-supplying systems, optimizing cooling systems to shrink the PUE index, and relying on buying more green energy as an alternative source. Another way out might be to wait for quantum computing to come into play.


Sources: nature