Platform

The EmberRTOS box you can set up with nobody on the phone

The EmberRTOS boot splash: the EmberNet logo above the word EmberRTOS on a white screen

Every edge box has a moment where somebody calls an engineer. The box is on a bench, in a panel, or on a trade show floor, there is no good way to reach it from the outside, and the person standing in front of it needs one answer. Which cable is which?

That answer usually travels over a bad phone line. So we put it on the screen.

What EmberRTOS is

EmberRTOS is our own operating system for the edge node. The root filesystem is read-only, every update lands as a snapshot, and a bad update rolls back instead of leaving you with a box you have to drive to. Machines stay identical, which is most of what you want from a machine you will not see again for a year. We wrote about the foundation in an earlier post, and the documentation covers updates and rollback in detail.

The part we want to show you is what happens at the machine. The node image boots, shows the EmberNet mark while it starts, and goes straight to a full-screen setup app on the box's own display. Nobody logs in and nobody connects in. If you can plug in a cable and look at the screen, you can bring the box up.

Six screens, in the order you need them

Network. Both wired ports are listed with their link state, speed, and address. Say which port the plant machines are on and which one the cluster uses. If you do not know which socket is which, press Blink this port and it flashes that port's light. We put that button there because the alternative is a guess.

The Network screen of the EmberRTOS setup app, listing both wired ports with link state, speed, and address

EmberNet. Make the install script for the box in EmberCORE, paste it in, and choose Enroll this box. If the box has no uplink yet, set up the cluster first and enroll later.

Cluster. The first box starts the cluster and shows a join code. The next two join as control planes, and every box after that joins as a node. Joining uses the code over the local network, not the dashboard, so it does not need an uplink.

The Cluster screen of the EmberRTOS setup app, showing the cluster running and its join code

Real-time. More on this one below.

Status. One screen that reads the box as it is right now: the kernel, whether the tuning is active, the watchdog, the cluster, whether the node is Ready, and whether anything has failed. It reads the machine when you open it, not a cached copy. The box in this screenshot has started a cluster but is not enrolled yet, so the EmberNet rows say so. The screen shows the box's real state.

The Status screen of the EmberRTOS setup app, showing the real-time kernel, the watchdog, and the cluster as Ready

Logs. When you need help, this builds a support bundle. It lists every file that goes in before anything is sent, and passwords, tokens, join codes, and keys are left out. If you are going to send a stranger a bundle from a plant, you should be able to read what is in it first.

The whole app is in English and Spanish, and one button in the corner switches.

The real-time switch

Some machines have a deadline, like a motion loop or a drive that has to be told what to do on time. Most do not, and making them carry the tuning a deadline needs costs you for nothing.

So the node image carries two kernels, and the real-time kernel is the default. The Real-time screen has one setting, On or Off. On runs the real-time kernel with its tuning: some processor cores are kept for real-time work, interrupts are moved off them, and deep sleep states are capped. Off runs the general-purpose kernel with none of that. A change takes effect after a restart, and a banner reminds you until you do. On a cluster, control planes restart one at a time, and a box will not restart while another node is not Ready.

What we have measured

We only quote numbers we measured ourselves. The one long run we have is a 24 hour soak on a Raspberry Pi 5, an ARM64 board, with a real-time kernel. The test ran at top real-time priority on three reserved cores while a heavy load, including a memory streamer, ran on the fourth core. The worst-case wake-up latency over the 24 hours was 23 microseconds. The load stayed off the real-time cores, so that shows the isolation working. It is not the worst case a machine can see.

We have not measured the x86 node image this post is about. The screens here come from it running in a virtual machine in our lab, in demo mode, so we are not quoting a number for it. When we have measured it on real hardware, we will publish the result and how we ran it.

Why we built it this way

Because the people who set up a box are rarely the people who built it. They are integrators, electricians, and plant engineers with a laptop that may or may not have the right software and a phone that may or may not have a signal. Every step that needs somebody at our end is a step that fails at the worst time, in a room we cannot see.

The Plasma firewall is on the box too, and it ships off in this version. One switch turns it on. The Status screen shows whether it is on.

If you run EmberNet, the setup guide walks through every screen, and the real-time guide explains what the two modes mean for your workload. The measured results and how we ran them are on the EmberRTOS page.