The Anatomy of an Edge Controller The reality is, none of these devices contain any parts that are truly proprietary or unavailable to the public. Part 4 of 4 - The Software Part 3 of 4 - The Hardware Part 2 of 4 - The Processor Part 1 of 4 - Introduction

Calvin Slater Member of Project Haystack  Project-Sandstar LinkedIn Contributing Editor

The Processor

Above we have the Sedona framework running on the Octavo OSD3358-512M-BSM. The Octavo chip is known as a System in Package or (SiP) houses a Texas Instruments AM3358 processor at 1GHz as well as 512Mb of RAM and full power management system, all integrated into a single device. In this small test, we have a 100-component Sedona DDC app and a single SOX connection. The app execution rate is set to ten milliseconds. This means that the DDC app will be scanned through 100 times every second, (Running the app at this frequency is somewhat of an extreme use case). At the end of every app cycle, the Sedona Virtual Machine (SVM) also processes any outstanding work required by all service components that are running on the platform (such as the SOX session). It does all of this from within the context of the SVM executable. After each engine scan, the app and SVM will go to sleep until the next scheduled scan cycle.

That last paragraph was an earful... Sorry. But it’s necessary to frame the topic we are about to discuss today; the Edge Processor. As you can see from this example above, the single Cortex-A8 core listed as “CPU” in the upper left corner is running at about 15.0 percent of its capacity. Of that total processing time, the “SVM process” which is listed as PID 1551 is consuming 20.7 percent of the core’s total computing efforts. The rest of the CPU’s time is committed to doing other things.

Those other things are all listed there as well. They each have a unique “PID” or Process IDentification. Every individual running process consists of one or more threads. A thread is a fundamental unit of code execution. It’s the path the CPU takes as it navigates through a particular piece of software; whether that involves completing a series of consecutive instructions, looping through the same code several times, or following some kind of fork in the road to another set of functions. We can see that there are currently 66 threads executing on our processor at this moment. It was mentioned before that the Sedona app executes and then goes to sleep. This is an example of a thread running and then blocking. Blocking means that the thread runs, and then pauses while it waits for some other event to occur. In this case, the event is the time until the next engine scan. Hopefully, our SVM got everything done in that 10-millisecond window and will go to sleep for the remaining period until it’s time to run again for the next cycle.

One important thing to know is that our CPU does not run this thread continuously until it blocks. In reality, it’s actually rapidly changing between all 66 threads, as long as they are also not blocked. It runs a little bit of code from one thread and then rapidly switches context to run code for another. This is the nature of an applications processor. These devices, which almost always use operating systems, switch rapidly between threads giving the illusion that many things are all happening at once. Processors that are designed to support operating systems with this kind of preemptive multitasking capability are what make a device an Edge Controller.

Contrast our CPU here to the action of a microcontroller. Microcontrollers are simpler processing devices with much less complicated hardware. The internal architecture is designed such that these devices would normally run a single thread only. In other words, they were designed to do only one thing. Sedona was designed to execute as a single thread exactly for this reason; so that it could be used on a low-cost microcontroller. At that time, that was all that was realistically available from a hardware cost perspective. This is no longer true today. Applications processors are now widely available and also inexpensive. So, if the microcontroller was good enough for what we were doing before, why would we move to use components that are much more complex? Why are all the major manufacturers of controls adopting these applications processors in their new devices? The answer is that these days we need our field devices to do more. The reason for this is the same reason all new smart devices use this type of hardware:

1. Increased bandwidth to support secure encrypted communications.
2. The ability to handle multiple applications, services, and connections at once.
   
3. Maintainability; the ability to apply firmware and security updates.

A programmable field controller has to be much more than it has been in the past: a flexible, secure, data-crunching, sensor hub that connects to anywhere in the world, rather than a slow, esoteric, plastic box, that mindlessly churns through the same setpoint code every day. So then, if the microcontroller cannot do all these things, is it dead for use in controls? Not at all.

[an error occurred while processing this directive] The virtues of the microcontroller are:

1. Lower power.
2. Deterministic code execution.
3. Fast booting and extremely reliable.

Consider for example the case of a smart Electric Meter. We all know AC current takes the form of a sinusoidal wave. Here in North America, that waveform is of course 60Hz. To capture information about electrical usage, we must continuously capture sensor samples at regular intervals. Per the Nyquist theorem, we must sample this wave at a minimum of at least 120Hz. In fact, we must sample at an even greater frequency to have any chance to be accurate. Comparing this requirement to our Sedona engine scan example from above, the 100Hz SVM scan frequency, even though extreme for the Sedona world, isn’t even close to what is needed. We would have to move this particular sampling task to a completely different processing context. Not only that, but the main CPU core we are using is not even appropriate for this job.

The Cortex-A8 application core is designed for high throughput and not determinism. That is to say, it can do a lot at once, but cannot guarantee when any of it gets done. Our edge processor cannot fail to take a sample at the correct time interval, or else all sensor accuracy will be lost. For that matter, it also can’t miss that SVM engine scan every 10ms. Care must be taken to give the Sedona process the highest priority in the thread scheduler. The real-time sampling use case is an area where the microcontroller core dominates. Because its hardware is simpler, it’s better at doing just one thing predictably, and reliably, and at regular intervals. It would be best to offload these kinds of tasks to microcontrollers and leave the less time-critical tasks to the apps processor.

Or even better, what if both processing cores could somehow be tethered to together on a single device and share certain resources? The application core would run the main OS and is responsible for PC or server-like duties, while the microcontroller core(s) are independently dedicated to single real-time measurement and control processes. Fortunately for us, this has already been thought of. This particular chip, as well as many others, have multiple cores! The AM3358 is actually a multi-core device. It has a single ARM core, a graphics processor, plus two microcontroller cores known as Programmable Real-time Units (PRU). These additional microcontrollers were included specifically for performing real-time processing independently from the main CPU core. This is very different than a desktop PC having multiple cores. Personal computers have multiple cores to distribute the workload for the same types of tasks. Embedded Edge Processors have multiple cores to specialize in specific tasks. This is known as Heterogenous System Architecture.

This high level of silicon integration is what is fueling the revolution of digital controls. Much of the groundwork has been laid out by the SOC developers as well as the software engineers who have ported their operating systems to these devices. Consider this recent announcement from the Linux Foundation regarding support for open-source RISC-V processor cores. The infrastructure supporting these tiny devices has been created and is being continuously supported. It took a tremendous effort by many to pack so much value into a single silicon wafer.

Whatever cannot fit on the applications processor die itself, such as RAM and flash memory is being packaged into ever-shrinking modules, such as our OSD3358 above. Devices such as these take advantage of recent advancements in packaging technology, in addition to all of the silicon integration we have seen. This tiny, postage-stamp-size chip is a fully integrated processing system. Which is another way of saying it’s a full PC on a single chip. The only thing its lacking is a screen, keyboard, and mouse. If you needed to add one, you could. Instead, we will be gluing this device to an Edge Controller baseboard. Hope to see you next month when we will talk about that topic in The Hardware.