Edge computing is all the rage now. The idea that the petabytes of data generated by industrial processes or even things like autonomous cars will travel to the cloud and back has always been in doubt. Issues associated with latency, bandwidth costs and security are just too much to overcome when lives and trade secrets are on the line.
We’ve seen this play out in several ways in the last few months. AWS Greengrass emerged and Microsoft’s IoT Suite added the ability to link edge devices back to the cloud. What’s most interesting is this idea that computing will need to happen in layers in the internet of things. Some compute will happen at the sensor layer where a cluster of related sensors provide a basic insight into the status of a process.
There will likely be a gateway layer where sensor data is combined with other machine data to determine if a machine is about to malfunction or needs attention. There might also be a computer vision layer that takes in camera or people tracking data to share with other sensors. In a car, this might cause the vehicle to brake. On a machine floor, it might cause a process to be shut down if a person wanders too close.
Once you accept that computing and analytics will happen at the edge, then we have to ask what those devices need. FogHorn Systems believes it has an answer. The company, which was formed in 2014, makes software that performs data analytics on as little as a few kilobytes of memory.
The company has built a complex event processing engine that’s akin to something like Apache Spark, only instead of running across many different compute nodes in the code, it’s designed to run on a sensor or a gateway. David King, CEO of FogHorn, says that when the company launched its product the tech world was still trying to push this idea of everything heading to the cloud for analysis. But by designing its own event engine from the ground up for resource-constrained devices, FogHorn is proving that industrial customers can keep their data close by and still get the intelligence they want.
One customer, an unnamed elevator manufacturer, can run 60 different machine learning models on the FogHorn software. That’s pretty impressive since the elevator maker has integrated the software into its own control system, not in a server that is running a cluster of graphics processors. Another customer is GE, which uses FogHorn’s software as part of its offering to industrial clients.
GE’s implementation represents King’s hope for his business: Selling the software to large systems integrators in the industrial, smart cities and intelligent enterprise. From there, the event processing engine could become a de facto standard. FogHorn has recently signed partnerships with GE, Bosch, Dell, SAP and others to ensure its software works on their gear and is included in their designs.
As computing moves to the edge and data are processed in different layers outside of the cloud, the question then becomes what are the new software stacks needed at each level, or perhaps in specific industries? That’s where we’re going to find the new leaders of the industrial internet of things.