- Primary cluster
- Compute clusters
Primary cluster is required. Compute clusters are optional.
You can put data in Block Storage on the primary cluster or in object storage.
However, to balance work across the different components, consider putting large and frequently-accessed medium tables in object storage, where compute clusters can access them. Object storage consists of the Object File System and external object storage access using Native Object Store functionality.
The following sections explain when the optimizer uses each component.
Primary Cluster
The primary cluster owns the data dictionary, parses and optimizes queries, controls transactions, and enforces workload management rules, such as concurrency limits. Every query starts on the primary cluster.
When a query accesses only primary cluster tables, the optimizer builds a plan that runs the query on the primary cluster.
When a query joins small primary cluster tables to large object storage tables, the optimizer plan is likely to run the larger steps on compute clusters.
A primary cluster table supports a primary index (PI) definition, but an object storage table does not. Therefore, applications that support a high volume of single-AMP requests run more efficiently when they access only primary cluster tables.
For a well balanced, cost-effective database, restrict Block Storage (primary cluster storage) to small tables and data needed to support primary index lookups or very short critical queries.
Related: Primary Cluster
Compute Clusters
Compute clusters are optional, have no permanent storage, and are for compute-intensive work. They are not part of the primary cluster. They can start and stop when the workload changes. While they can process data from any table, they can directly access only object storage tables.
For effective compute cluster use, you must put at least one table in the query in object storage.
Related: Compute Clusters