Azure Cosmos DB: How Pricing, Data Organization Works

Azure Cosmos DB, a service that provides globally distributed databases that can support multiple database modes, employs several concepts for organizing data and charging for deployment activity. The items described here apply to the provisioned throughput deployment option for all supported database modes except PostgreSQL.

(For an illustration, see “Azure Cosmos DB Containers and Partitions.”)

Provisioned Throughput Pricing Model

Cosmos DB’s provisioned throughput pricing model charges customers based on a Microsoft-created capacity unit called a Request Unit (RU), which is derived from an unpublished formula that factors in compute, memory, and IOPS usage. Customers purchase a provisioned throughput rate (RU/s) for each container. Requests are denied by the service when traffic surpasses the provisioned throughput rate purchased.

The RU/s rate is distributed by the service evenly across the physical partitions in a container, although a feature in preview allows customers to specify the throughput assigned to each physical partition. Customers who choose the provisioned throughput pricing model can minimize costs by reducing the required RU/s for their deployment, and the recent scaling features described in the main report assist in this effort.