Compute Clusters | VantageCloud Lake - Compute Clusters - Teradata Vantage

Teradata® VantageCloud Lake

Deployment
VantageCloud
Edition
Lake
Product
Teradata Vantage
Published
January 2023
Language
English (United States)
Last Update
2024-02-17
dita:mapPath
phg1621910019905.ditamap
dita:ditavalPath
pny1626732985837.ditaval
dita:id
phg1621910019905

Compute clusters exist as separate clusters of BYNET-connected nodes independent of the Primary Cluster.

Each compute cluster is separate and may be specialized. Compute clusters enable you to extend and automatically scale up compute power.

Compute clusters don’t have any permanent data. A data dictionary structure exists on a compute cluster, but it serves only the transient needs of the compute cluster. It does not contain table or column descriptions or details about statistics, indexes, or privileges. All that detail is in the primary cluster.

A compute cluster can read large tables in object storage. It can also hold intermediate data, keeping it in memory or in internal drives.

Compute clusters are for queries that access Object Storage to perform compute-intensive work such as analytic functions. This frees up the primary cluster to perform session management, parsing engine, and tactical or other short-term work.

Compute clusters are 100% fault-isolated from system-wide failures. They are independently upgradeable to newer versions of software without disrupting the rest of the system.

How Compute Clusters Work

You may have zero or many Compute clusters in your Environment. It depends on the workload. Compute clusters are independent, so you can add or remove one or more compute clusters as needed. Groupings of compute clusters can automatically scale up additional compute clusters based on resource demand or the number of active queries.

One query step can run on only one compute cluster. The optimizer in the primary cluster determines which query steps go to a compute cluster and builds a query plan. A plan may include processing a step or two in the primary cluster, and then processing one or more steps on a compute cluster. This varies from query to query.

All queries start in the parsing engine of the primary cluster. Communication between the primary cluster and compute clusters occurs through Query Fabric, which passes data and instructions about how to process it. The primary compute router directs a request from the primary cluster to an active compute cluster. The router is aware of which clusters are active and which are on standby. It establishes a session with a cluster based on routing rules.

The compute results go back to the primary cluster for return to the client application using XML Open Database Connectivity (xODBC) or native Command Line Interface (CLI) connectivity.