PowerCenter Integration Service Architecture Overview

The PowerCenter Integration Service moves data from sources to targets based on PowerCenter workflow and mapping metadata stored in a PowerCenter repository. When a workflow starts, the PowerCenter Integration Service retrieves mapping, workflow, and session metadata from the repository. It extracts data from the mapping sources and stores the data in memory while it applies the transformation rules configured in the mapping. The PowerCenter Integration Service loads the transformed data into one or more targets.

To move data from sources to targets, the PowerCenter Integration Service uses the following components:

•PowerCenter Integration Service process. The PowerCenter Integration Service starts one or more PowerCenter Integration Service processes to run and monitor workflows. When you run a workflow, the PowerCenter Integration Service process starts and locks the workflow, runs the workflow tasks, and starts the process to run sessions.

•Load Balancer. The PowerCenter Integration Service uses the Load Balancer to dispatch tasks. The Load Balancer dispatches tasks to achieve optimal performance. It may dispatch tasks to a single node or across the nodes in a grid.

•Data Transformation Manager (DTM) process. The PowerCenter Integration Service starts a DTM process to run each Session and Command task within a workflow. The DTM process performs session validations, creates threads to initialize the session, read, write, and transform data, and handles pre- and post- session operations.

The PowerCenter Integration Service can achieve high performance using symmetric multi-processing systems. It can start and run multiple tasks concurrently. It can also concurrently process partitions within a single session. When you create multiple partitions within a session, the PowerCenter Integration Service creates multiple database connections to a single source and extracts a separate range of data for each connection. It also transforms and loads the data in parallel.