Rocman Components

Rocman contains five types of key components: top-level iterations, agents for physics modules, actions, schedulers, and coupling schemes.

• One of the major tasks of Rocman is to drive the simulation. For this purpose, it provides top-level iterations including time-marching schemes for both steady and unsteady-state calculations. In the driver code, Rocman invokes time integration of the coupling scheme by passing in the current time and obtaining a new time, until the system reaches a designated time or a converged state.

• An agent serves a physics module. It represents a domain-specific simulation (fluid, solid, or combustion) in a coupling scheme. The most basic task of an agent is to initialize the physics module and manage its persistent buffer data for use during intermodule interactions on behalf of the physics module.

• Interactions between physics modules are encapsulated in actions. An action is a functional object implementing a designated calculation. An action also defines the input data, on which it operates and the output data produced by the calculation.

• A scheduler is a container of actions, and is responsible for determining the orders of initialization, execution, and finalization of its actions. A scheduler provides a procedure add_action() to its user for registering actions. After all the actions have been registered with a scheduler, the scheduler can then automatically schedule these actions based on the data flow among actions. The automatic scheduling constructs a call graph, which is a directed acyclic graph (DAG) for the actions, in which each edge between a pair of actions is identified by the data passing from one action to the other. This automatic scheduling of actions greatly simplifies the work of an end-developer, who now needs to be concerned about only the data movement among actions without having to worry about the order of its execution. Furthermore, constructing a call graph of actions exposes parallelism among actions and potentially enables concurrent execution of all independent actions that have their input data ready. In the future, we plan to extend the run-time scheduling to allow concurrent execution of actions.

• A coupling scheme is composed of a number of agents and a scheduler. The scheduler determines the orders that must be followed for invoking initialization, execution, and finalization of agents and actions. The coupling scheme is the only code an end-developer of a new coupling scheme needs to write. Rocman provides a rich set of predefined basic actions, which can then be used as building blocks for new coupling schemes.