The IBM PPL Library contains the following main components:

• Runtime

The parallelism model employed is task based. In this model a parallel computation consist of a set of tasks that are executed by the scheduler. For load balancing the scheduler uses work stealing. To better exploit parallelism in various applications two levels nested parallelism is allowed. This means that from the main thread you can create a parallel computation as a collection of tasks with possible dependencies between them and each task in turn can create a parallel computation as another collection of tasks. The tasks at the second level for now can't create additional work though this limitation may be removed in the future.

For the user the library provides for each, map, map reduce and schedule task graph as main functions to express parallelism.

• DataStructures

The graph is the main data structure supported for now. We implement by default an adjacency list with support for directed, undirected graphs, properties for vertices and edges and support for various storage classes.

• Messaging

To operate in a distributed environment both the runtime and various datastructures employ a message API that can be impleemnted on top of a native messgaing library.

• Applications

The library also contains a set of examples and full applications for graph transformations and bayesian .