System Requirements

Before installing Koios, verify that your host machine meets the following requirements. If you have not already, Installation Overview lays out the whole path and what to gather first.

Koios runs all services inside a single Docker container — the configuration database, time-series database, in-memory cache, web proxy and the application services — and the figures below account for that.

Hardware

If you have no idea yet, start at 4 cores, 8 GB and 100 GB. That comfortably runs a few hundred tags with a model or two, and it is a sensible first guess for a site you have not surveyed.

The floor column is where Koios still starts and runs. It is not where it runs well, and a machine built to the floor has nowhere to go when the site grows.

Memory and CPU can usually be raised later in a few minutes — on a virtual machine it is a settings change and a reboot. Disk is the one to be generous with, because growing it means resizing the virtual disk and then the partition inside it, on a machine people are relying on. It also costs nothing to over-provision: a dynamically expanding virtual disk consumes only what is actually written.

Processor Architecture

Koios runs on 64-bit Intel and AMD processors (amd64) and on 64-bit ARM (arm64), which covers Raspberry Pi and most industrial ARM boards. A 32-bit operating system is not supported on either.

If you are unsure which a machine has, run uname -m on it: x86_64 is amd64, and aarch64 is arm64. It matters when downloading an offline archive, which is published separately for each.

Running Models

Models are the main reason to sit at the upper end of the CPU and memory ranges. A handful of small models running at a slow cadence barely registers; large models, wide inputs, or inference on every scan cycle all want more of both. A small ARM board is well suited to data collection and modest inference, and will struggle with large models.

Stacks

If your components run in stacks — isolated package sets built from wheels you upload — budget for them separately:

• Memory and database connections. Each stack in use runs one worker process, whatever number of component environments are attached to it. Budget roughly 200 MB of memory per stack, plus its own pool of connections to the configuration database. Reusing one stack across several component environments costs far less than giving each environment its own.
• Disk. Built stacks and the wheels uploaded into them are stored alongside your other uploaded files. A single wheel can be up to 500 MiB and an imported bundle up to 2 GiB, so an install using several stacks can consume many gigabytes beyond the historical data above.

Operating System

20.04 LTS still runs Koios, but it is past its standard support window and no longer receives ordinary security updates — plan to move off it.

Ubuntu Server (minimal install) is recommended. A desktop environment is not required. Koios is managed entirely through its web interface.

Software

Docker is the only software dependency. See Installing Docker Engine for setup instructions.

Network

On a virtual machine, use bridged networking so Koios can reach devices on the local network. Choosing a Host has the settings for each hypervisor.

What's Next

• Choosing a Host: decide between a virtual machine and a physical computer