Could Millions of Smart Meters Be Used to Create a Powerful Supercomputer?
Hive Computing wants to rival the processing power of Google and Amazon by harnessing idle smart meters.
Jeff St. John
August 7, 2015
Smart meters can do a lot more than count kilowatt-hours. We’ve been tracking how utilities and advanced metering infrastructure (AMI) vendors are putting them to use to detect power outages, connect home energy devices, and even serve as nodes of a distributed computing architecture for the grid edge.
In fact, the communications cards in most modern smart meters have roughly the same computing power as a cellphone, Angel Orrantia, co-founder of startup Hive Computing, says. Most of the time, they’re sitting idle — and that represents a huge computing resource.
Hive wants to put that processing power to use for tasks ranging from cracking cancer’s genetic code, to forecasting the long-range weather impacts of climate change, Orrantia explained in an interview last week. Over the past year, the company has written the software to farm out computing tasks that meters could work on when they’re not busy, “built on the premise that a lot of these communications modules are relying, basically, on smartphone processors.”
Hive won’t be asking these mini-computers to solve problems that require low-latency, high-bandwidth movement of data between them, because AMI networks aren’t designed to do that, he said. Instead, it will focus on problems that can be solved by asking millions of relatively simple questions, where “it’s a small payload that comes in, it’s a small payload that comes out — but in computer time, it’s there for quite awhile,” he said.
This kind of thing has been done on desktop PCs for years, he noted. The University of California, Berkeley’s SETI@Home project has enlisted some 3 million people to allow their computers to be used to process radio telescope data in search of extraterrestrial intelligence, for example. The Berkeley Open Interface for Network Computing (BOINC) has created open-source code to allow spare PC capacity to be similarly volunteered for all sorts of socially useful computing tasks.
Of course, Hive plans to charge users, since it wants to pay the utilities for the privilege of borrowing their AMI computing power. In that sense, Hive would be competing against cloud computing providers such as Amazon Web Services, Microsoft, and Google, which make unused servers available for similar computing tasks, Orrantia said.
“They have to build their infrastructure for peak usage, and when it’s not at peak usage, they can sell that compute time to people who take advantage of it. We’re doing the same thing,” he said.
“The difference is, we don’t have to pay for the underlying hardware, we don’t have to cool it, and we don’t have to worry about the underlying power consumption,” he continued. Any electricity used by smart meters’ comms cards is essentially lost in their overall power consumption profile, and is too small to lead to the circuit heating issues that require active cooling for server farms, he said.
That means that Hive should be able to offer its computing service at a much lower cost — about $200,000 for 30,000 cores for one month, compared to nearly $2 million or more for the equivalent from Google, Microsoft or AWS, according to Hive’s calculations.
“The initial market we’re looking at right now is for cancer treatment — developing the genome for various cancers, so that the retrovirus can be created,” he said. The cost of computing power today forces cancer researchers to limit the number of analyses they do on this front to a limited number of “best guesses,” said Orrantia, but “with this computing power, they wouldn’t have to guess anymore — they could just run the full model, and take the human error out of the equation.”
Moving the smart-meter-as-supercomputer from concept to reality
All of this presumes, first, that Hive can actually tap that latent smart-meter compute, and second, that it will get the permission needed from vendors and utilities to do so. On the first front, Orrantia said that there’s little reason why the firm shouldn’t be able to work with most communications modules in smart meters today.
“We’re building our platform right now with the intent that we’re going to deploy on legacy equipment,” he said. The main requirements are that they have to be running on Linux, since Hive has designed its software on a Linux kernel. They also have to have a reasonable communication channel that doesn’t impose additional costs.
That latter point means that cellular-linked meters might not work, since they come with data charges. But mesh networks run by utilities don’t come with extra costs, and would be ideal for Hive’s low-bandwidth data needs.
On the second front, Hive is starting to test it out on communications cards from smart-meter vendor Itron. “They’ve already provided us with some of their hardware, so we can start porting our solution from the lab onto real meters,” said Orrantia.
Itron has been working with partner Cisco to embed Linux-programmable intelligence into its latest generation of smart meters, which works well for Hive, since it has designed its software to run on a Linux kernel.
It’s also working with Glen Canyon, the startup that’s quietly delivering millions of $25-and-under smart meters to utilities in China, India and other emerging markets, he said. It’s hoping to roll out its first real-world test in Glen Canyon meters being deployed for a smart city project in Beijing, under the auspices of gigantic utility State Grid Corp. of China, although “that’s a few years out, because they’re still in the planning stages.”
John Heibel, CEO of Glen Canyon, said his company is volunteering all the meters they can use in their development. “Since our stuff is over-the-air downloadable, it’s easy for them to do. I hope they make a go of it — that will find at least one great commercial application for smart meters,” said Canyon.
Hive also has the connections that could help it get entry to those meters. Startup co-founder Eric Frazier has worked for Cisco, Silver Spring Networks, and Itron acquisition SmartSynch. The company has also landed some key metering communications experts, such as former Itron and Cisco IPv6 mesh network developer Jun Zha and grid communications cybersecurity expert Elias Gonzalez.
Smart meter and grid communications vendors are already hard at work trying to open their platforms to more complex computing tasks. Itron’s Riva platform, Silver Spring’s SilverLink Sensor Network, and Elster’s Connexo platform are some of the most prominent examples of the work underway on this front.
As for utilities, Orrantia noted that San Antonio, Texas municipal utility CPS Energy is interested in Hive’s potential. Raiford Smith, CPS’ vice president of corporate development and planning, is the former chief of giant utility Duke Energy’s emerging technologies group, where he led the so-called “Coalition of the Willing” effort to turn smart meters into nodes for a distributed intelligence platform for grid control.
Duke has since brought together two dozen grid and IT vendors to design a set of interoperable, standards-based technologies for allowing field devices to work together in the field, and CPS Energy is planning two projects to test the technology. While the partners in this work have been thinking for some time about how to tap smart meters’ latent computing power for utility tasks, “we weren’t thinking of complex algorithms for cancer genome research,” Smith said.
“Like most utilities, we’re very interested in things that look to expand or extend our investment, and make the business case look better,” said Smith.
CPS isn’t yet working with Hive. It would have to carefully vet its cybersecurity plans to make sure it doesn’t compromise utility assets, and make sure the way it puts those meters’ comms cards to use doesn’t interfere with utility data-collecting operations.
But as Smith has pointed out in the past, the latent capacity of the world’s smart meter network approaches that of the world’s better-known supercomputers. For example, 3,000 smart meters have nearly the same amount of processing power and memory capability as Deep Blue, the IBM supercomputer that beat Garry Kasparov in a game of virtual chess in 1997, and 150,000 meters add up to about half the computing power of IBM’s Watson supercomputer, he said.
With millions of smart meters able to do work, “This would be like flooding the market with supercomputing capacity, if it’s real,” he said. “This would make the price of what Amazon and Google are doing plunge.”