When I first heard about fog computing, I laughed. If ever there was a marketecture, it was fog. Just another buzzword meant to tag along behind IoT and try to drum up some media coverage. And then a couple of years later, Foghorn turned up, wanting to brief us. Fair enough. We’re game.
The idea of fog computing is that IoT devices stream so much data, that it’s impractical to shunt it all to the public cloud for processing. IoT data must be analyzed locally in real-time to be of value in many applications, particularly industrial ones.
Besides tending towards high latency, bandwidth to the public cloud is simply not cheap enough to size pipes sufficiently large for industrial IoT applications. Thus, the term “fog” was coined to conjure an image of a cloud that’s close by, rather than one far away. Sending data into the local fog allows for speedy analysis and timely results. No pesky bandwidth delays while shuttling terabytes or petabytes of data streaming off of your IoT devices.
Foghorn’s play is to locally process the huge amounts of data generated by industrial IoT (IIoT) machines. By “industrial,” we mean mining, jet engines, automated manufacturing, large refineries, large retail shops, and smart meters, by way of example.
And if you think that sounds like shipping data over the LAN to a nearby data center, think again. Foghorn challenges this traditional fog computing paradigm somewhat, bringing “intelligence at the edge.” Lots of IIoT devices don’t have local data centers in which to perform data processing. And even if they do have a data processing center closer than the public cloud, connectivity is often fragile.
How does Foghorn work?
Foghorn’s local data processing is as local as it gets. A device running dual core x86 with typically 2-4GB of RAM is assigned per IIoT device. These devices are small — as small as a cigarette pack or as large as the battery used to power a golf cart.
Data streamed from the IIoT device hits the locally attached Foghorn box, and is processed immediately using code written in VEL, Foghorn’s own programming language that creates “virtual expressions.” Foghorn’s CEO David King described VEL as reminiscent of SQL, but optimized to handle real-time data ingestion and to work effectively with the time-series database that’s resident entirely in RAM.
After the data is processed, high-value metadata — the results of what was processed — is shipped upstream to a local aggregation point. There, the data can be monitored via a console and acted upon. The whole process is meant to be incredibly fast, enabling real-time decision-making by software or good old-fashioned human operators.
Reducing the data set to processed metadata is key, as this reduces the data that must persist by a factor of 100x to 1,000x, depending on the application. That makes it plausible to use public cloud for data persistence once the initial processing is over.
Who’s going to use Foghorn?
Foghorn’s investors include a number of interests, including GE, Bosch, Siemens, and Rockwell, although many other companies are involved. These sorts of investors are to be expected, considering the primarily industrial use cases.
Foghorn described a number of real-world applications, including the following.
- Wind energy forecasting, where wind turbine data is processed by Foghorn to accurately predict power yield for the next 24 hours.
- Factory yield optimization, where temperature and pressure sensor data is analyzed to discover a manufacturing facility problem that would result in a bad component being produced.
- Cavitation alerts, where temperature, input pressure, output pressure, and water velocity are monitored in real-time to detect the conditions in which an air bubble might be introduced into the system. These air bubbles, or cavitations, can damage or destroy expensive water pumps.
- Locomotive fuel efficiency, where local engine sensors are coupled with additional data such as GPS to reduce engine idle time, saving significantly on fuel. Even at idle, locomotives ingest an enormous fuel load. Reacting to fuel overuse meaningfully requires real-time data analysis.
If you’re interested in a Foghorn deployment, you’ll likely be working with a global systems integrator and/or your equipment vendors. While it’s possible to program Foghorn devices as an end-user, you might be hard-pressed to devote the staff.
For more on Foghorn, visit http://foghorn-systems.com/.