Skip to main content
This document describes the hardware requirements and sizing guidance for running Omni on-prem. Omni itself has relatively low resource requirements. In most environments, disk performance (primarily for etcd) and storage capacity for images are more important than raw CPU. The numbers below are practical starting points based on internal guidance and etcd recommendations. As always, test with workloads representative of your environment before production use.

​
Deployment models

A typical Omni on-prem deployment includes:
  • Omni
  • etcd
  • (Optional) Image Factory
  • (Optional) Local container registry
For more details, see the Run Omni On-Prem documentation. In small and medium environments, all components can run on a single VM. For larger deployments, you can separate components onto multiple machines.

​
Full stack on a single VM

If you run Omni and all supporting services (etcd, image factory, and container registry) on a single VM, the following is a reasonable baseline for up to ~200 managed nodes:
  • 4 vCPUs
  • 8–16 GB RAM
  • 500 GB SSD storage
  • 1 Gbps network (2.5 Gbps recommended for heavy WireGuard traffic)
For environments with fewer than 50 managed nodes, you can typically provision about half of these resources:
  • 2 vCPUs
  • 8 GB RAM
  • 200–250 GB SSD storage
These estimates assume moderate usage and that image caching is enabled.

​
Component breakdown

The following sections describe the resource requirements and sizing considerations for each component individually.

​
Omni

Omni is lightweight. Scaling from tens to hundreds of managed nodes does not significantly increase its CPU or memory requirements. For small and medium deployments, Omni can safely share a host with etcd, as long as etcd’s storage requirements are met and the disk provides good performance.

​
etcd

Omni uses etcd to store cluster state. Disk performance is critical for stability and responsiveness. General guidance (based on etcd recommendations):
  • CPU: 2–4 cores for typical clusters
  • Memory: 8 GB is sufficient for most deployments
  • Disk: Fast SSD strongly recommended

​
Disk performance

etcd is sensitive to disk write latency. Slow disks can lead to:
  • Increased request latency
  • Leader election instability
  • Overall cluster performance degradation
Recommended minimums:
  • SSD-backed storage
  • At least ~500 sequential IOPS for moderate workloads
  • Higher IOPS for large or heavily loaded environments
Avoid slow spinning disks whenever possible. If HDD must be used, choose the fastest available option and validate performance using disk benchmarking tools. RAID 0 can be used to improve disk throughput. When running three or more etcd members, additional RAID mirroring is typically unnecessary, since etcd already provides replication at the application level.

​
Container Registry

If you are running a local container registry for Talos and Kubernetes releases:
  • Plan for 3–4 GB per Talos and Kubernetes release
  • Include additional space for extensions and system containers
  • Size storage based on how many versions you plan to retain
For example, retaining 10 releases requires approximately 30–40 GB at minimum. Allocate additional headroom for future growth.

​
Image Factory

Storage requirements depend on:
  • The number of images you build
  • Whether caching is enabled
In most setups:
  • Fewer than 10 installation images are maintained
  • Images are cached to avoid rebuilding during installs or upgrades
In practice, 1–2 GB of total storage is typically sufficient for a single Omni deployment. Allocate additional space if you frequently rebuild images or support multiple image variants.

​
Network requirements

For most deployments:
  • A 1 Gbps network connection is sufficient
  • Maintain low latency between Omni and etcd
  • Keep components in the same data center whenever possible
For larger environments or deployments with heavy WireGuard traffic:
  • Use 2.5 Gbps or higher to improve throughput