8 tips for building a cost-effective IoT sensor network

iot saving money

With more than 2 million sensors already installed, and facing an expansion plan for smart metering that would add millions more, Thames Water needed to make its data network infrastructure as flexible, reliable and secure as possible without driving costs up.

The company, which provides drinking water and wastewater services to more than 15 million customers across London and the Thames Valley, had to rethink its entire approach to networking, ranging from protocols to how collected data would be stored and analyzed.

Thames Water joined with consulting firms Accenture and Deloitte, the Bilfinger engineering company and IBM to form the Technology and Transformation Alliance to help it create new ways of working and delivering technology services and projects. A key part of the plan is upgrading Thames Water's existing monitoring network to exploit advanced analytics.

"What I’ve inherited is a set of infrastructure that’s been implemented over many years," says Simon Coombs, an Accenture managing director who is leading Thames Water's Intelligence Hub network advanced analytics program. "You have a mix of high-density customers in the middle of the capital city and you have, of course, lower density through the countryside -- a wide range of different issues to contend with as a company."

One way Thames Water hopes to achieve a more cost effective sensor network, without sacrificing performance, reliability or security, is by gradually replacing older standards with newer Internet of Things (IoT) protocols.

"The IoT is providing the opportunity to bring much cheaper devices in," Coombs says. Yet for enterprises like Thames Water that have operated massive sensor networks for many years, the road to improved cost efficiencies is pitted with the challenge of supporting a range of different protocols. "Of course utilities want to reduce cost, but they also have a portfolio of predominantly old kit that they’re having to still manage, and maintain and deal with," Coombs says.

Most of Thames Water's existing sensors, which monitor variables such as flow and pressure, are wired devices connected to local remote telemetry units. "The remote telemetry unit supports a range of protocols, but the majority, because they're relatively old, are using GSM or ADSL," Coombs says. "So it's regular, old-fashioned broadband."

Data from the telemetry system is captured and organized by OSIsoft’s PI System, where it can be accessed by employees or sent to the cloud for further analysis. By mining these real-time data feeds, Thames Water has been able to increase water quality while lowering energy costs and also reduce the number of pollution incidents.

Thames also turned to Amazon Web Services (AWS) and Microsoft Azure. Carrying out advanced analytics in the cloud provides a way to quickly allow access to data using new and old protocols to communicate effectively during what promises to be a lengthy transition period.

"New cloud techniques are allowing us to connect, directly down into particular devices to intercept them at the lower protocols," Coombs says. "We’ve got much more flexibility now in the cloud, and that flexibility is allowing us to drive down costs."

Moving to the cloud has also helped Thames Water reduce costs by eliminating the need to manage network operations from a conventional brick and mortar data center. "In the olden days when you had to build your own infrastructure it could take months, years to build solutions," Coombs says. "Now we can very quickly, within a matter of days or weeks, run up new approaches on the cloud, and new analytics streams, and we can test and deploy them very quickly."

Here are some tips for deploying an IoT sensor network, culled from interviews with Coombs and others who have gone down this road.

IoT tips

1. Don’t overspend on unnecessary features

While no two sensor networks are identical, and enterprises have different needs and goals, a variety of basic management approaches and strategies can be used to bring costs under control.

A fundamental cost efficiency rule is to select technology that meets a project's current and future requirements without wasting money on unnecessary features or maintenance.

"For example, if you choose a Power Over Ethernet (POE), sensor versus a battery-operated wireless sensor for a high frequency sensor application, then you eliminate the need for battery maintenance and Wi-Fi interference," says Morne Erasmus, global building solution lead for CommScope, a network design firm headquartered in Hickory, N.C.

"On the other hand, a short term deployment might benefit from magnetic attachable, battery operated, Wi-Fi sensors," he adds.

2. Clearly define sensor roles and capabilities

Placing too many different types of sensory nodes into a network without clearly defining their roles and needs is another mistake many enterprises make. "Throwing bandwidth at these devices before doing any testing or knowing their capabilities is a waste of money and resources," says Destiny Bertucci, network monitoring head geek for SolarWinds, an Austin, Texas, firm that provides IT monitoring and management tools. "Network administrators should know specifically what they need from a sensor device in order to get the most out of it."

3. Use testing and monitoring tools

Once a network is up and running, many enterprises waste money by failing to use testing and monitoring tools to accurately track exactly how much network bandwidth is actually being consumed at specific times. "By having an idea or baseline of the usage, you can have a better understanding of how adding more [bandwidth] will affect the overall cost of your network," Bertucci says.

4. Don’t be afraid to mix and match

Like Thames Water, enterprises planning a large sensor network should take a close look at integrating existing components and services into their current network infrastructure, says Matt Leipnik, founder and CEO of Chalk Circle, a network consulting company based in Kent, England. "The biggest costs for a sensor network are initial physical deployment of sensors, ongoing maintenance and management at scale," he notes.

Coombs says that enterprises shouldn't be afraid to mix old and new sensors and proprietary and open protocols. "That used to be scary, but we now say that's fine," he says. "Connect to it and then put your standardization up a layer in the real-time data layer," he advises. The idea, Coombs explains, is to create an optimal environment "where you're using that data as opposed to worrying too much about the standardization of the individual devices."

5. Transfer sensor data at off-peak times

Deploying sensors capable of both receiving commands and transmitting data to a central management location can lower costs by making data transfers more flexible and efficient, such as by sending information only when it is actually required. "Certain times of the day may require more horsepower than others, but having the ability to dynamically manage sensor networks in the same way that mobile operators are able to throttle their network back and forth with demand is incredibly important because otherwise you have inefficiencies," says Ryan Martin, a senior analyst at ABI Research.

6. Only send data when something changes

Enabling selective data collection from a massive number of sensors can help prevent "big data" from turning into "gargantuan data," driving up both data transfer and storage costs without adding much extra insight. "If you're collecting this information at, say, six-second intervals every day throughout the year, not just for one facility but a bunch of them, not just in one country, but the world, that can be pretty messy," Martin says.

"The reality is that not all that data is necessarily required to be collected for some analysis to occur." Bertucci agrees. "At the end of the day, having data just to have it is a waste of time, storage and IT resources," she says. "Too much monitoring results in just noise, and the actual metrics--the ones that provide actionable insights--get lost."

To reduce unnecessary data traffic, Simon Jordan, head of industrial sensing at technology consulting firm Cambridge Consultants, says that enterprises should consider building networks that pre-analyze data directly at the sensor. Only data that has changed, or looks interesting in some important way, needs to be sent to the data center or cloud. "We term this 'Edge Intelligence'," Jordan says. "It treats the sensor, the network and the analytics as one."

Jordan says that massive sensor networks should be viewed as layers of agents that filter and forward data based on relevance rather than just buffering or translating between protocols. "Think of the difference between an army's chain of command and a post office," he says. "Soldiers can make small decisions locally and pass important information back up the chain, whereas the mail just passes every letter on," he explains. "This not only prevents the classic information overload, but reduces the load on the communications."

7. Balance cost and performance

When building efficient sensor networks, enterprises need to think about cost and performance tradeoffs, as well as geographic dispersion and required reliability, says Brandon Davito, vice president of smart cities and streetlights for Silver Spring Networks, a smart grid products manufacturer headquartered in San Jose, Calif. "It’s really a matter of understanding the network requirements against those different axes, then selecting network platforms that are built on open standards that have the security and the flexibility to provide an upgrade path for now and into the future as these applications become more and more mission critical," he explains.

8. Start small

Finally, companies that address sensor network deployment via a crawl-walk-run approach tend to achieve the greatest cost efficiencies, says Mark Benson, CTO of Exosite, a Minneapolis company that develops IoT software. "Key to building a cost-effective sensor network that scales in an efficient manner are [organizations] that start small, learn quickly and grow over time, all while ensuring that they are focusing their measurement efforts on the parameters that are closely correlated with finding anomalies and trends that make an impact on the business."

John Edwards is a freelance writer. He can be reached at jedwards@johnedwardsmedia.com.

Copyright © 2017 IDG Communications, Inc.

The 10 most powerful companies in enterprise networking 2022