How the Open Compute Network works

A job submitted to the Open Compute Network is divided into cells — units of between thirty seconds and fifteen minutes of work. Each cell is identified by a SHA-256 over its code, its inputs and its parameters, so identical work has an identical identity anywhere and at any time. The network prices the cell against a published tariff, checks the cache, selects a machine that can run it, and verifies the result before money moves. Settlement is the last step and the ledger row is immutable.

A cell is the unit of scheduling, pricing, verification, caching and re-running.
Cell identity is a SHA-256 of the image digest, input references, parameters, schema version and runtime class — nothing about where or when it ran.
Cached results are served without dispatch and without charge for the computation.
A budget is a hard ceiling: the network stops rather than exceeds it and returns partial results.

Elsewhere on this site

The Cube Floor — The Open Compute Network. The live figures on this page are loaded from https://api.thecubefloor.com, which is public and needs no credentials to read.