7.00.02 - Node Types - Aster Execution Engine

Aster Instance Installation Guide for Aster-on-Hadoop Only

prodname
Aster Execution Engine
vrm_release
7.00.02
created_date
July 2017
category
Installation
featnum
B700-5022-700K

There are two types of nodes in an Aster instance: queen nodes and vWorker nodes. For more information about the Aster instance architecture, refer to the Teradata AsterĀ® Execution Engine Aster Instance User Guide.

Queen Nodes
The queen is the central node in the system, and is responsible for managing the execution engine cluster configuration and coordinating queries. It is installed on a Hadoop cluster edge node. Responsibilities of the queen node include:
  • Accept, optimize and plan SQL or SQL/MR statements from clients
  • Distribute work across the vWorkers in the cluster
  • Aggregate results from workers
  • Send results back to clients
Vworker Nodes

Vworkers are the daemons that are distributed over participating Hadoop cluster data nodes and are responsible for computation on distributed data. A vWorker is the unit of parallelism in the execution engine working on a slice of the input data independently and coordinated by the queen. As your computational needs grow, you can add more vWorkers to the execution engine cluster to ensure query execution times scale linearly with data volume and computational needs. Each participating Hadoop data node hosts one or more (typically four-twelve) vWorkers. The vWorkers receive phases of the execution plan that are relevant to a slice of the distributed data. Data that reside on HDFS is read in parallel by the vWorkers via the Aster-Hadoop connector. Each vWorker is also responsible for the storage of a slice of Aster transient data in its local temporary storage space. Vworkers have the capability to perform inter-vWorker communication in case the plan requires shuffling of data between worker machines. Once the computation is complete, the results are sent back to the queen.

The number of vWorkers in the cluster determines the computational capacity of the execution engine. The engine capacity can scale to match required need by adding more vWorkers. Adding vWorkers can be achieved by increasing the number of vWorkers per data node (up to the maximum number of vWorkers per node recommended), or by increasing the number of participating Hadoop data nodes. The latter increase may require adding new data nodes to the Hadoop clusters if all data nodes are already participating.