Services

Bare metal for compute that is uneconomic in the cloud, managed as code

When you need it

Data-heavy workloads with fast local disks, such as ML experiments, data processing or build farms, cost disproportionately much in the cloud, but the team avoids "hardware" because nobody can reproduce it.

What you get

A cluster on rented servers (Hetzner or similar) managed as code: installation and bootstrap with Ansible, RAID and LUKS according to your data requirements, object storage access without keys on the hosts, a check on the availability of the needed server configurations at the provider, basic observability. A documented procedure for adding, removing and reinstalling a node and a guide for the cluster's users. Handover: your team adds the next node by following the procedure.

Typical duration

A pilot of 2–3 nodes: 2–3 weeks. Growth to around ten nodes and handover: another 2–4 weeks. Estimate after a first call.

Related work

Work example 3.

Contact