IBM Streams 4.2

Developing streams processing applications with user-defined parallelism

The @parallel annotation allows developers to easily take advantage of data-parallelism. In the streaming context, data-parallelism means replicating copies of operators, and splitting streams so that different tuples go to each set of replicas. The process of replicating operators, and creating all of the new streams to connect them, is called the parallel transformation. When developers add @parallel to a primitive or composite operator invocation, IBM Streams will perform the parallel transformation at submission time. The goal is to improve overall application throughput by executing the replicas in parallel.

  • User-defined parallelism
    User-defined parallelism, available through the @parallel annotation, allows you to easily take advantage of data-parallelism in your IBM Streams applications.
  • Setting the degree of parallelism
    You can specify the number of channels for parallel regions within an application or as a submission time value.
  • Parallel region restrictions
    While the @parallel annotation can be applied to most operators, there are some limitations. The following conditions are not allowed in parallel regions, and the compiler will generate an error if it encounters them.
  • Parallel transformations and fusion
    Parallel regions are expanded when you submit a job. The expansion process is called the parallel transformation, which transforms the logical version of the application produced by the compiler into the physical version of the application which can be deployed to the distributed runtime.
Parent topic: Writing streams processing applications
Related tasks:
Validating the topology of your job in Streams Studio