Holoscan Core

View as Markdown

An Application is composed of Fragments, each of which runs a graph of Operators. The implementation of that graph is sometimes referred to as a pipeline, or workflow, which can be visualized below:

Core concepts: Application
Core concepts: Application
Core concepts: Port
Core concepts: Port

The core concepts of the Holoscan API are:

  • Application: An application acquires and processes streaming data. An application is a collection of fragments where each fragment can be allocated to execute on an exclusive or shared physical node of a Holoscan cluster.
  • Fragment: A fragment is a building block of the application. It is a directed graph of operators. A fragment can be assigned to a physical node of a Holoscan cluster during execution. The runtime execution manages communication across fragments. In a fragment, Operators (Graph Nodes) are connected to each other by flows (Graph Edges).
  • Operator: An operator is the most basic unit of work in this framework. An operator receives streaming data at an input port, processes it, and publishes it to one of its output ports. In applications executed by holoscan::gxf::GXFExecutor, an operator corresponds to the GXF-specific Codelet concept.
  • Port: A port is an interaction point between two operators. Operators ingest data at Input ports and publish data at Output ports. Port is a Holoscan concept used by both holoscan::gxf::GXFExecutor and holoscan::GPUResidentExecutor. In holoscan::gxf::GXFExecutor, ports map to the GXF-specific Receiver, Transmitter, and MessageRouter concepts used to move data between connected operators.
  • Executor: An executor that manages the execution of a fragment on a physical node. The framework provides executor implementations such as holoscan::gxf::GXFExecutor and holoscan::GPUResidentExecutor to execute an application.
  • Scheduler (C++ (holoscan::Scheduler)/Python (holoscan.core.Scheduler)): A scheduler decides when each operator in a fragment runs. The SDK ships three — GreedyScheduler, MultiThreadScheduler, and EventBasedScheduler. EventBasedScheduler is the recommended default for pipelines with parallelism. See Choosing a Scheduler for the decision table and configuration guidance.

GXF Executor-specific Concepts

The following concepts are specific to applications executed by holoscan::gxf::GXFExecutor:

  • Codelet: A GXF-specific unit of execution. In the Holoscan SDK, this role is represented by an Operator.
  • (Operator) Resource: Resources such as system memory or a GPU memory pool that an operator needs to perform its job. Resources are allocated during the initialization phase of the application. This matches the semantics of GXF’s Memory Allocator or any other components derived from the Component class in GXF.
  • Condition: A condition is a predicate that can be evaluated at runtime to determine if an operator should execute.
  • Message: A message is a generic data object used by operators to communicate information.

Holoscan 4.1 uses FlowGraph and FlowGraphImpl for the existing flow-oriented graph API, along with the Python module holoscan.flow_graphs.