Data Centers, Cooling and AI
Cooling in Data Centers
Aritifical Intelligence (AI)
That Artificial Intelligence is on the rise is an understatement. We are seeing high impact data science applications on a daily basis within very different fields.
Machine learning enables large data sets to be analyzed quicker than ever. Algorithms write new algorithms and improve – literary – overnight.
IBM’s Deep Blue won over Garry Kasparov, AlphaGo by Google’s DeepMind topped the Korean grandmaster Lee Sedol at Go. Recently AI poker bot Libratus, developed by Carnegie Melon University, has beaten four top poker players at heads-up, no-limit Texas hold’em– a game of “imperfect information”, where it is extremely complicated to figure out the ideal strategy given every possible approach your opponent may be taking.
While those seem like pop-culture cases, these cases are the laying foundation for business AI. One of the most publicly debated cases right now is autonomous driving.
Car manufacturer Tesla is leading the way into a driverless future. Vehicles will be equipped with self-driving hardware, which relies heavily on machine learning in order to achieve a safety rating substantially higher than that of a human driver.
Furthermore, cooling, heating and ventilation is the next frontier ready for disruption by AI and machine learning.
Energy Use in Data centers by the Numbers
Keeping data centers cool as they crunch data is a massive challenge and the amount of energy consumed by big data centers has always been a headache for tech companies.
In 2014, U.S. data centers used 70 billion kWh of energy – equal to powering more than six million homes for a year.
This is an important shift in energy consumption:
- From 2000 to 2005, usage grew 90%,
- From 2005 to 2010, usage grew 24%;
- From 2010 to 2014, usage grew 4%.
Energy use is expected to increase at the same rate of 4% from 2014 to 2020. (Source: Berkley Lab)
More efficient Cooling of Data Center with the help of AI
Facebook built one of its facilities on the edge of the Arctic Circle in order to lower its cost to operate a data center and improving its environmental friendliness.
An example of machine learning and big data improving lives as well as profits is the Google Data Centers Case. Here Google is putting its DeepMind artificial intelligence unit to manage power usage in parts of its data centers.
The result – according to Google – is a 40% reduction in the amount of electricity needed for cooling after applying DeepMind’s machine learning.
After accounting for “electrical losses and other non-cooling inefficiencies,” this 40% reduction translated into a 15% reduction in overall power saving.
Google described it “a major breakthrough”; considering it has been achieve on a software level only with no improvement of the hardware.
How did they do it?
DeepMind researchers improved the system’s utility by: “using a system of neural networks trained on different operating scenarios and parameters within our data centres, we created a more efficient and adaptive framework to understand data centre dynamics and optimize efficiency.”
They took historical data previously collected by thousands of sensors within the data center.
Data like:
- temperatures,
- power,
- pump speeds,
- setpoints and many more
They used it to train an ensemble of deep neural networks.
They trained the neural networks on the average future power usage effectiveness (PUE).
PUE = is defined as the ratio of the total building energy usage to the IT energy usage.
Then they trained two additional ensembles of deep neural networks to predict the future temperature and pressure of the data center over the next hour.
The purpose of these predictions was to simulate the recommended actions from the PUE model, to ensure that they do not go beyond any operating constraints.
After that, they have deployed the model on a live data center. The graph below shows a typical day of testing.
The curve significantly decreased when machine learning (ML) recommendations were turned on, and significantly increased when machine learning (ML) recommendations were turned off again.
A graph showing how using machine learning helped Google reduce power consumption in its data centers. (Image credit: DeepMind)
Large-scale impact
With this machine learning system Google is able to consistently achieve 40% reduction in the amount of energy used for cooling. What is even more ground breaking is that this algorithm is a general-purpose framework (opposed to a narrow-purpose framework, which is the case with autonomous driving). This general purpose AI is largely more capable and applicable in solving complex problems and can be applied to other challenges in the data center environment and beyond.
The impact becomes even more significant, if you consider that Google has in total 12 enormous data centers all over the World where there are more than 900,000 servers. To put this into perspective:
Google’s data centers use around 260 million watts of power which accounts to 0.01% of global energy (which is the equivalent of 200,000 average homes being consistently powered).
In addition, Google has dozens of others facilities where it could apply this algorithm to. And what is even more interesting, the Google’s DeepMind team will roll out the system broadly and share the know-how, so other data centers and industrial system operators might use it as well.
If we combine this with Menerga’s Air Handling Unit Adcoolair 75 for data centers that allow for free indirect cooling and adiabatic evaporative cooling, the savings are going to be even greater.
Green engineering can even go one step further. One example is the positive engineering practice of cooling with the help of sea water. The most efficient Google’s data center facility is in Hamina, Finland. The high-tec cooling system uses sea water from the Gulf of Finland in order to reduce energy use. This approach is similar to Menerga’ s sea water cooling of hotels which are located at the beach or close to the sea.
Machine learning and Business – the not-so-far future of any company
If you haven’t yet, start the exploration of emerging practices and technologies as soon as possible – on a full-team scale. Literary, any department can benefit from it. Some of them will be an obvious and indispensable ingredient in your future-commpany.
Lead and take the first step today. Gather a team of data science professionals who are passionate about asking the right questions and identifying problems worth solving.
The Davids who are going to beat the Goliaths are the people who understand that data will lead the way. Your future-company will be thankful for that.
Sources:
https://eta.lbl.gov/publications/united-states-data-center-energy-usag
https://www.google.com/about/datacenters/efficiency/internal/
https://deepmind.com/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-40/
https://www.cloudyn.com/blog/10-facts-didnt-know-server-farms/
https://www.theverge.com/2016/7/21/12246258/google-deepmind-ai-data-center-cooling
http://www.datacenterknowledge.com/archives/2012/10/17/how-google-cools-its-armada-of-servers
