As we navigate the internet, we don't usually think much or realize how our queries send a request to a server in a data center to retrieve the content.
There are approximately 8,000 data centers in the world, and with the rise of social, mobile, analytics, and cloud services worldwide, it is expected a growth rate of 10.9% a year by 2030.
With the growth comes a cost. Data Centers have a proportionally big footprint.
They account for approximately 1% of the energy consumed worldwide. The largest existent data center could power 80,000 U.S. households. The primary consumers are the cooling system, the IT system, and the electrical system.
Depending on the energy source (solar, coal, wind, or other), it will contribute more or less to the emission of greenhouse gases.
- IT equipment data, like server, storage, and telco equipment
- IoT sensors on switches, routers, generators, batteries, HVAC,
- Physical security systems, like biometric and video surveillance
- Staff information
- Maintenance information
- Energy consumption
Predictive Maintenance
On the other hand, predictive maintenance is when something is detected that indicates a failure could happen if nothing is done. This is possible using analytics and artificial intelligence that monitor the health of the assets and detect when some anomaly happens in the equipment.
Most equipment already has some kind of analytics, like detecting if a specific sensor went above or below some pre-specified threshold.
Artificial Intelligence can detect more complex patterns, like correlation between multiple variables and unexpected pattern changes in time series sensors. Data centers have thousands of sensors, and simple analytics cannot identify these complex patterns.
Failure prediction and anomaly detection are the most common techniques in this case. However, a failure prediction is a supervised approach; therefore, it needs labels. In other words, it needs many failure records to learn. This is only sometimes possible since, in situations like this, managers prefer to perform more maintenance than needed and prevent failure before it happens, as the costs of failure could be catastrophic.
Anomaly detection is an unsupervised approach. It learns the patterns of the signals and will assign an anomaly score for each sample analyzed. The facility manager will receive an alert if the score is too high. Deciding the right threshold for the alarm is the biggest challenge because managers don't want the system to alert for every small change in pattern that happens, but also they don't want to ignore when something needs to be taken care of. With appropriate maintenance, not only does the system become more reliable, but the life of the equipment is also extended.
Energy Consumption Forecast and Optimization
As mentioned above, data centers consume a tremendous amount of energy. Reducing consumption can significantly decrease the amount of greenhouse produced by energy generation.
Intelligent analysis of equipment usage can detect equipment that could be used more efficiently and help in the creation of plans for energy consumption improvement, like implementing different cooling techniques and upgrading equipment that is energy efficient.
When able to accurately forecast energy consumption, it is easier to plan and match renewable energy production with the data center's needs, reducing the amount of fossil fuel utilization.
Besides, servers running below their capacity could be consolidated with other servers and air conditioning could be optimized according to the number of IT equipment, people in the building, humidity, airflow, and other variables.
Tasks that consume more energy can also be scheduled to run during off-peak hours, optimizing workload distributions.
Predict Building Energy Consumption using AutoAI
Comments
Post a Comment