Moving compute and storage resources to edge locations can reduce latency and bandwidth needs, improve performance and save money. At the same time, widespread edge computing deployments can introduce significant management challenges. Servers can be hard enough to maintain when they’re in an on-prem data center. What if they’re deployed in the middle of nowhere?
Energy companies know all too well the challenges of remote computing.
“When we drill a well, it’s always in the middle of nowhere,” says Dingzhou Cao, senior advisor for data science at independent shale producer Devon Energy, a Fortune 500 company based in Oklahoma City, Okla.
Sending a massive stream of data back to a central location isn’t always feasible, but the company still wants to know what happens at its sites. “We always have a bad connection to the Internet,” he says. “Ninety percent of the time it’s available—but 10 percent of the time, it’s not.”
Ninety-percent availability is acceptable when the data is for monitoring purposes only, but if a real-time response is required, it’s a problem. In particular, the company is looking to improve drilling and operations efficiency and automate tasks with machine learning. For that, on-site team members need immediate data analysis.
“We want to move everything to the edge, so that even if we lose the connection, the guy at the site can still see what’s going on and make a decision,” Cao says. “If you rely on cloud computing and lose the connection, then the field guy can’t see the results.”
To make it happen, the company faced the challenge of putting, in effect, small data centers out in these far-flung locations.
Skills availability was an early concern. “At the drill site, they’re not computer guys,” Cao says. “They’re not IT guys. It’s too much to expect them to set up the system for you, so the simpler the better.”
Prioritizing ease of deployment and management
The company chose Hivecell’s edge-computing-as-a-service platform. Hivecell provides the hardware—bright yellow, stackable modules—and the cloud-based platform that allows customers to configure and manage the hardware and software.
HivecellThe Hivecell boxes snap together magnetically if more than one is needed, either for added computing power or for redundancy. Connecting the modules together requires no cabling. Instead, Hivecell’s patent-pending Baranovsky connectors pass both power and data between the modules. The entire stack requires just two cables: one power cable, one ethernet cable. According to the company, there are no limits to how many boxes can be stacked on top of one another.
Each three-pound module comes with a 64-bit ARM processor with six CPU cores at 2.4 Ghz and 256 GPU CUDA cores. In addition, there’s 8GB of memory and a 500GB SSD. For networking, a Hivecell supports gigabit ethernet to an external network, with two additional gigabit ethernet channels for module-to-module communications. There’s also built-in WiFi in case there’s a temporary separation from the network, and, in case of a power outage, 25 to 60 minutes of battery backup, depending on workload.
The management system allows customers to schedule and monitor the rollout of Hivecell nodes to dozens or hundreds of locations, deploy containers with a push of a button, upgrade the operating system and Kubernetes runtime at all locations remotely, and set threshold alerts for Hivecell clusters and control performance in real time.
Devon Energy installed the Hivecell modules right in drilling rigs or in data vans at the hydraulic fracturing site to collect data from the equipment, says Cao. To get each unit up and running, a Devon employee plugs it into a power outlet, connects the serial cable (Hivecell added a serial to Ethernet adapter specifically for Devon Energy), and turns it on. The Hivecells connect back to the corporate network via LTE or satellite connections, depending on service availability at each specific location.
“It’s a one-button installation,” says Cao. “And we rent the hardware, so if one of the Hivecells isn’t working, they send me a new one.”
Cao says he likes Hivecell’s ease of deployment and management, and for him, “It’s the best design in the market right now.” Plus, it was a complete package. “Other competitors either sell you the hardware, or they sell you the software.”
Cao started evaluating edge-computing options at the start of 2020, back at WPX. WPX completed its merger with Devon Energy this January. The company started deploying the platform in late 2020, and has four Hivecell boxes out in the field. The first two were deployed at two hydrofracking locations in Texas, and the third was deployed at a fracking location in North Dakota. “They’re running great,” he says.
The hardware runs Apache Kafka, an open-source real-time data-feed platform. In this first stage of deployment, the Hivecells are collecting data from field machinery.
The amount of data at the sites isn’t particularly high, Cao says. A new row of data, containing about 40 columns, is generated each second. “The volume of data is in kilobytes, not terabytes,” he says. “The benefit of edge computing for this use case is not the volume of the data. It is to process the data at the edge in true real-time, so that the analytics results can be fed back to the field engineer to aid decision making, without worrying about the internet connection instability and signal latency.”
There are different analytics that can be run at the site, he says, including hydraulic fracking event detection, automatically classifying the type of drilling that is being performed, predicting hydraulic fracturing pressure in real-time, optimizing hydraulic fracturing costs in real-time, and optimizing directional drilling paths.
Engineers can access the real-time data on their cell phones or computers. “We’re putting real-time data in the hands of the people who need it most,” he says.
“We aren’t running any [machine-learning] models yet,” he says. “We’re just collecting the data and setting up the foundation for the next phase.”
The data collection phase is expected to last until the end of June, he says. “We’re already in the process of building the [machine-learning] model, but we haven’t reached the stage where it’s ready for production. We plan to have it in the field by the end of this year.”
At that time, the engineers will also have real-time analytics that they can use in their wellsite decision making, and the Hivecells will also aggregate data and send it back to headquarters via the LTE or satellite connections.
Cao already anticipates how easy it would be to distribute the final machine-learning model from the cloud out to the edge Hivecells. “In the future, we’d just press a button and sync the model to the Hivecells.”
Edge computing across industries
It’s not just the energy sector investing in edge computing. According to Hivecell co-founder and CEO Jeffrey Ricker, the growth of IoT and 5G are spurring edge computing growth in many different industries. He’s seeing interest from communication companies looking to use edge computing to help with 5G rollouts, manufacturers looking to deploy machine learning on factory floors, retail companies, and shipping and logistics. “Our customers are looking to go to hundreds or thousands of locations, with multiple Hivecells at each one,” he says.
Hivecell is currently in paid pilots with 24 Fortune 500 companies, Ricker says. COVID-19 did put a dent in the startup’s launch, he admitted. “Hivecell went into full production in the fourth quarter of 2019, and we started doing pilots with customers in the first quarter of 2020,” he says. “Not the best time to launch a startup.”
As a result of the pandemic, customers put projects on hold for most of last year, he says. But in late fall, things started picking up again. “This quarter, we’ve been overwhelmed,” he says. “It’s been remarkable.”`
The company’s biggest challenge, Ricker says, was how to create a system that is easy to install—not just for the first server, but the second, third, fourth and fifth one. Because companies don’t want just a single server in the field, but several, for redundancy and performance reasons.
Hivecell wound up having to build its own hardware, he says, to create a system that could snap together and automatically form computing clusters. “Our goal is that if you can deliver a pizza, you can install a Hivecell box,” he says.
“Half of the edge projects being deployed this year will fail because they cannot scale,” Ricker says. “They may work in the lab. They may work in a dozen locations. But when you get to hundreds—it does not scale. So that’s where we’re focused – running clusters of servers at hundreds of thousands of locations.”
Copyright © 2021 IDG Communications, Inc.
