Datacenter Energy Consumption isn't an Autonomous Vehicle Problem, it's an Everything Problem

I caught a great conversation the other day between some of my favorite members of #energytwitter: oil experts Anas Alhajji and Jamie Webster, natural gas pro Richard Meyer, and Steve Levine of Axios.

They were discussing a sometimes overlooked aspect of connected and autonomous vehicles (CAVs), the massive amount of data involved. And more directly, the energy consumption required to handle the influx of data as CAVs begin to enter the market.

Years ago, I gave a talk on CAVs and I told the audience that when people think of these cars, the "driverless" part is usually the first thing that comes to mind. But the game-changing part about these vehicles isn't that they know how to drive. The true magic is the connected part, that your vehicle will also possess the knowledge of all the other nearby cars as well. The vehicles will communicate with each other constantly using vehicle-to-vehicle (V2V) communication. Imagine if you could (and your brain could handle) constant communication with every driver around you on a high-speed highway. Each car knowing when the other will slow down, speed up, change lanes, or how and when they will quickly react in emergencies. Armed with all of this information the "network of connected vehicles" will view the situation both as a challenge of getting their individual vehicle to its destination in the most efficient manner, but also as a giant fluid dynamics problem as they work to collectively smooth traffic and eliminate traffic jams.

The information sharing doesn't end there. There's also vehicle-to-infrastructure (V2I) and vehicle-to-cloud (V2C) communication as well. In the former, all types of things will also talk to your car. The on-coming street light is telling everyone about its timing patterns so that all the cars can proceed through the intersection efficiently. The parking space around the corner, suddenly available, communicates that information to all the parking-seeking vehicles around. Sensor-equipped roads will literally know the location of potholes and tell vehicles the best way to avoid them. In V2C, the vehicle will constantly communicate with the cloud to report on road repair conditions and learn new info about traffic miles ahead, upcoming weather patterns, and necessary map upgrades.

The data is all consolidated and processed by the algorithms in the vehicle. And the data isn't just used to get the vehicle from point-A to point-B, but also influences how the vehicle makes the trip and in a manner that's most efficient for the passengers and for vehicle energy consumption.

This is an unprecedented amount of data that will all live on our growing global collection of servers inside of datacenters. The energy consumption of our datacenters isn't something we normally think about, but it's significant. Right now datacenters in the U.S. account for 2.5 percent of U.S. electricity consumption, sufficient to power every household in New York City twice over. The explosive growth of the internet and the continued development of "cloud" data storage and processing will only expand datacenter energy use in the future. In fact, datacenter energy consumption is already expected to double in just eight years.

So it's easy to appreciate the concern over how CAVs will affect datacenter energy consumption, but it's also a far too narrow view of the situation. In the coming Internet of Things (IOTs) revolution, vehicles are just one of those "things." CAVs aren't driving this trend, they're just along for the ride.

What's the solution to the energy stress caused by this flood of data over the next decades? The solution is not stopping the development of CAVs or the IOTs - and it's doubtful that we could do either, even if we wanted too.

The solution is two-fold: generation and efficiency. First, more electricity demand means generating more electricity, preferably from renewable and advanced nuclear sources. It also requires an advanced electrical grid better able to handle greater input from intermittent sources.

Second, we simply have to have better datacenters and increased research to those ends. There are a number of ways to go about this, but one path has to do with the interconnect technology used to transmit information between devices within a datacenter.

When an data request reaches a datacenter, it has to travel from server to server in order to find the proper information. Each one of these "jumps" between devices consumes energy. Currently, metal interconnects are used to transmit this information using electrons, but this process has limits on the amount and speed of data transmitted. New research is examining photonic interconnects, interconnects that use light photons, instead of electrons, to transmit data. If developed, the technology can send more data, at faster speeds, and with less energy used per bit of data. It will also allow for flatter network design, which would reduce the number of "jumps" required to access data. Widespread implementation of photonic interconnect technology could double the energy efficiency of datacenters.

The IOT data revolution is coming whether we like it or not, regardless if it comes from autonomous vehicles or from the thousands of other IOT devices. The only question is if we can handle the load with the proper combination of new post-carbon electricity generation and advanced highly-efficient datacenters.