Google fights rising data center power costs with AI
- Google's data center cooling systems are now directly controlled by cloud-based artificial intelligence systems developed by the company and Deepmind, its AI subsidiary. The solution is driving energy savings of 30% on average and is the first of its kind, according to a company announcement last week.
- The AI system pulls a snapshot from thousands of sensors in the cooling system every five minutes and sends it to a deep neural network, which analyzes the data and predicts courses of action to minimize energy consumption.
- This process reduces the manual load on operators, and a list of safety constraints in the local control system provides redundancy to avoid actions that might jeopardize the data center's safety and operations.
When Google started using machine learning to maximize data center efficiency, algorithms told engineers to "shut down the entire facility" in order to conserve energy. But after fine-tuning and tweaking, the company reduced its energy consumption and increased reliance on renewable energies, thereby also reducing its energy footprint.
The costs and complications of data centers and the relative simplicity of the cloud have driven companies to infrastructure as a service providers for full or hybrid cloud architectures. By handing the upkeep over to a provider, businesses can focus their efforts on platforms, software, data and other IT priorities.
For IaaS providers and data center management companies, this means more critical business infrastructure sits on their hardware — and wiggle room for problems and downtime is nonexistent. Cybersecurity is one of the most imminent external threats, with the potential of three to six days of downtime ringing in at least $15 billion.
Internally, the costs for maintaining data centers are rising. Gartner estimates that increases in demand and cost per kilowatt-hour are driving a subsequent 10% or higher increase in power costs for businesses. Data center efficiency needs optimized cooling mechanisms, which businesses can achieve with the installation of economizers, the containment of heat and quipment and air conditioning systems, according to Gartner.
Microsoft has taken to the seas with Project Natick to sink its data centers off of the Northern Isles and test the feasibility of underwater systems. The project launched its second phase in June, though commercialization still appears several years away.
But for the data centers already in place and companies that don't want to mess with hard-to-access locations, AI solutions can offer short and long-term solutions for energy consumption.
Follow Alex Hickey on Twitter