TTK is Getting MPI-Ready

Area of Science:

• Scientific Visualization
• Computational Topology
• High-Performance Computing

Background:

• Existing topology-based approaches for distributed-memory environments often use tailored, single-algorithm implementations.
• There is a need for versatile frameworks supporting complex topological analysis pipelines in parallel.

Purpose of the Study:

• To document the technical foundations for extending the Topology ToolKit (TTK) to distributed-memory parallelism using Message Passing Interface (MPI).
• To introduce a versatile approach for topological analysis pipelines supporting triangulated domains and regular grids.

Main Methods:

• Extension of TTK's data structures for triangulation representation and traversal to support MPI.
• Development of an intermediate interface between TTK and MPI at both global pipeline and fine-grain algorithmic levels.
• Classification of distributed-memory topological algorithms based on communication needs and examples of hybrid MPI+thread parallelization.

Main Results:

• Demonstration of a versatile approach for distributed-memory topological analysis pipelines.
• Performance analyses showing parallel efficiencies from 20% to 80% with negligible overhead from MPI-specific preconditioning.
• Successful application of the extended TTK to a 120-billion-vertex dataset on a 1536-core cluster.

Conclusions:

• The TTK extension provides a robust and versatile framework for large-scale topological analysis in distributed-memory environments.
• The developed methods address algorithmic and software engineering challenges in parallel topological computing.
• A roadmap and recommendations are provided for future development of TTK's MPI extension.