How-to
Install licensed software on an industrial edge PC
Short answer
Install vendor-licensed container software on an industrial edge PC by passing four gates in order: a purchased license with the installer obtained only from the vendor, a verified Docker engine, an installer build that matches the CPU, and a proven open-internet path before the pull. Vendor guidance budgets the install command at 30 minutes to 1 hour; at a 5 Mbit/s link a 2 GB payload alone needs about 53 of them.
On this page
- Gate 1: a purchased license and a vendor download link
- Gate 2: a working Docker engine on the host
- Gate 3: the installer build that matches the CPU
- Gate 4: an open internet path, proven before the pull
- Run the install and log in
- Worked example: pull time for a 2 GB application on a 5 Mbit/s link
- Questions asked
Installing vendor-licensed container software on an industrial edge PC comes down to four gates, passed in order: a purchased license with the installer taken only from the vendor, a verified Docker engine, an installer build that matches the CPU, and a proven open-internet path before the pull starts. Each gate has one pass reading; do not begin the next until the current one reads pass. Vendor guidance budgets the install command itself at 30 minutes to 1 hour, because it pulls every container over HTTPS — at a 5 Mbit/s uplink, a 2 GB application alone accounts for about 53 of those minutes. The steps below fit any edge PC on a supported 64-bit Linux distribution; only the build you download and the vendor CLI's subcommand names change from one product to the next.
Gate 1: a purchased license and a vendor download link #
The entitlement is per instance: it is purchased before installation, and each installed instance needs its own, so one license never covers a fleet of edge PCs. The installer download is issued by the vendor against that purchase, or against an accepted evaluation — vendors run pilot programmes for exactly this case, and the vendor's sales or support office sets one up.
Pass reading: you hold both the entitlement and a vendor-issued download link for the installer. Failure action: there is no legitimate install path without them. Complete the purchase or request evaluation enrolment, and take the installer from nowhere else — an unofficial copy cannot be licensed and is unverified code on a plant machine.
Gate 2: a working Docker engine on the host #
The installer pulls and runs containers, so a Docker engine must exist on the host before anything else. Run:
docker --version
A version string is the pass reading; use docker version when you want the client and server detail. Command not found means installing Docker Engine first: take it from Docker's own repositories, following the installation page for your distribution — Docker publishes the supported list there, along with which edition's license terms fit your use on that operating system.
Size the host against the vendor's published minimums at the same time; one industrial edge platform, for instance, documents 16 GB of RAM, a 300 GB disk and 4 CPU cores for its management environment. If you are standing the host up new instead of reusing a cabinet box, a fanless industrial mini PC sized to those minimums is the class of machine this install expects. This is the same engine and the same version check as when deploying Edge Data Store in Docker on a controller PC, so a host already running that job is through this gate.
Gate 3: the installer build that matches the CPU #
Installers of this kind ship in two builds: amd64 for 64-bit x86 processors and arm64 for ARM. Read the CPU on the device with uname -m: x86_64 takes the amd64 build, aarch64 takes the arm64 build. Download that build from the vendor link — the architecture is part of the package name — and move it to the edge PC over SSH:
scp installer-file user@edge-ip:/home/user/
Answer yes to the host-key prompt and enter the password; the entry stays hidden while you type. Then open an SSH session on the device, run cd ~ and ls: the installer file must appear in the listing, and every later command spells its name exactly as listed. A No such file or directory reply on the permission or run step means the typed name and the listed name differ.
Pass reading: the installer is on the device, listed by ls, with a suffix matching the CPU.
Gate 4: an open internet path, proven before the pull #
The installer fetches its payload over HTTPS and pulls containers from the vendor's registry, so at least one Ethernet interface needs an open or DMZ route to the internet. Prove it on the device before running anything:
ping -c 5 example.com
Replies mean name resolution and routing both work. A Temporary failure in name resolution reply splits one of two ways with ping 8.8.8.8: if the address answers, routing is fine and DNS is broken — correct the DNS server the interface receives, by DHCP option or static setting, and retest. If that second ping returns Network is unreachable, the interface has no route, which usually means no lease: an address beginning 169.254 in the device's network settings — 169.254.243.206 is a typical reading — confirms the interface never received a DHCP lease. Restart the network stack with sudo systemctl restart systemd-networkd on distributions that use it, wait a few seconds, and re-ping.
Errors at install time map to the same faults:
| Install-time error | Meaning | Action |
|---|---|---|
dial tcp: lookup download-host on 127.0.0.53:53: server misbehaving |
The resolver cannot resolve the vendor's download host: no working internet or DNS path | Connect an interface to the internet and redo the ping checks above |
tls: failed to verify certificate: x509: certificate signed by unknown authority |
An inspecting or filtering device on the path presents a certificate the edge does not trust | Move the interface to an open or DMZ segment, re-ping, and retry |
Temporary failure in name resolution |
DNS fault or routing fault | Split the two with ping 8.8.8.8 as above |
Network is unreachable |
Interface unconfigured or no DHCP lease | Read the interface in the network settings and restart the network stack |
Keep that internet-facing interface off the machine network for the install window — layered OT security controls are what guard the route once it opens — and change the application password at first login.
Pass reading: example.com answers by name and 8.8.8.8 answers by address on the same interface.
Run the install and log in #
With all four gates read pass, the install is a short fixed sequence:
- Add execute permission:
sudo chmod +x installer-file.ls -lthen shows an x in the permission field. - Run the installer:
sudo ./installer-file. An error here maps to the table under Gate 4; anything outside that table goes to the vendor's support. - Confirm the CLI: run its help command and read the subcommand list. Command not found means the installer did not finish — repeat step 2.
- Configure: run the configure step and enter what it prompts for, an edge identifier and a customer identifier; both are free-form names you choose.
- Install: run the install step. It pulls every container and takes 30 minutes to 1 hour; keep the session open for the whole pull, because a dropped connection can abort it.
- Start: run the start step to launch the pulled containers.
- Status: run the status step, which must report the application running. A stopped or partial state means running start again before moving on.
- Log in: open
https://edge-ipin a browser. The self-signed certificate warning on the first visit is expected — proceed past it to the login page. Sign in with only the credentials the vendor's manual documents, and change that password before the PC touches the plant network.
Decision: status reads running and the application's web screen shows its version — the install is done. Anything short of that loops back to the gate whose reading you skipped.
Worked example: pull time for a 2 GB application on a 5 Mbit/s link #
Inputs, each from documentation rather than guesswork: the payload is 2 GB, the per-application container size one industrial edge platform states in its system requirements; the link rate is 5 Mbit/s, the I/O budget the same requirements table grants such an application and a realistic figure for a constrained DMZ uplink; the documented install window is 30 minutes to 1 hour for the vendor's install command.
The rule in words: transfer time in seconds equals payload bytes times 8 bits per byte, divided by the link rate in bits per second.
The arithmetic: 2 GB = 2,000,000,000 bytes; 2,000,000,000 × 8 = 16,000,000,000 bits = 16,000 Mbit. At 5 Mbit/s: 16,000 ÷ 5 = 3,200 s, and 3,200 ÷ 60 ≈ 53 minutes. On a 100 Mbit/s plant link the same payload takes 16,000 ÷ 100 = 160 s, about 2.7 minutes.
Result: at 5 Mbit/s, pure transfer consumes roughly 53 of the 60 budgeted minutes; at 100 Mbit/s, transfer is negligible and the rest of the window is registry latency and image extraction.
Decision, by the link rate: a 5 Mbit/s link sustained for one hour moves 5 × 3,600 = 18,000 Mbit, against a 16,000 Mbit payload — it fits the documented hour with about 2,000 Mbit to spare, which is exactly why the budget is an hour. Book the full hour on a link of that class and do not interrupt the pull. On 100 Mbit/s or better, book 30 minutes. If your proven link rate times 3,600 s comes to less than the payload in Mbit, the pull cannot finish inside the window — fix the path under Gate 4 before you start.
Questions asked
How long should the container install take on an edge PC?
Vendor guidance for this class of installer budgets 30 minutes to 1 hour for the install command. The transfer arithmetic explains it: a 2 GB payload is 16,000 Mbit, which a 5 Mbit/s link moves in about 53 minutes and a 100 Mbit/s link moves in under 3. Keep the SSH session open, or run the install inside a terminal multiplexer, because a dropped connection can abort the pull partway.
Can I evaluate the software before buying the license?
Most industrial software vendors run an evaluation or pilot programme for exactly this situation, so ask the vendor's sales or support office about enrolment before planning the install. The installer download arrives only with a purchase or an accepted evaluation, and the entitlement is per instance, so a single license does not cover a fleet of edge PCs; each installed unit needs its own.
What if the edge PC has no Docker engine?
Install Docker Engine before running the vendor installer. Docker's own documentation lists the supported distributions, including 64-bit Ubuntu releases such as the 24.04 LTS, and states which edition fits which use, so take the engine from there rather than from an arbitrary package source. Verify with docker --version, which should print a version string, and only then move on to the installer download.
The web login rejects the documented credentials after a restart. What now?
A restart error can leave the application half-configured, which shows up as a rejected login with credentials you know are correct. Re-run the vendor CLI's configure step, enter the edge and customer identifiers again, then sign in with the credentials the manual documents and set a new password. If the reconfigure does not restore access, hand the case to the vendor's support line.
Can the edge PC install the software with no internet access at all?
Not with this class of installer. It fetches its payload over HTTPS at install time and pulls every container from the vendor's registry, so at least one interface needs a proven open path before you start. An offline install exists only where the vendor documents an offline bundle for the product, so ask the vendor rather than improvising one.