Ensuring domain consistency in an adaptive framework with distributed topology for fluid flow simulations

TL;DR

This paper presents a decentralized domain management strategy for fluid flow simulations on large parallel computing clusters, reducing communication bottlenecks and supporting adaptive domain topology changes.

Contribution

It introduces a hierarchical data structure and communication patterns that ensure domain consistency without global broadcasts, enabling scalable adaptive simulations.

Findings

Decentralized management reduces communication overhead.

Supports adaptive refinement, coarsening, and migration.

Ensures domain consistency in large-scale simulations.

Abstract

Top-tier parallel computing clusters continue to accumulate more and more computational power with more and better CPUs and Networks. This allows, especially for environmental simulations, computations with larger domain sizes and better resolution. One of the challenges becoming increasingly important is the decomposition and distribution of the overall work load. State-of-the-art parallel codes usually use solutions that involve complete knowledge of the domain topology, which will lead to communication and memory bottlenecks when computing very large domains. To meet this challenge, the authors propose a new strategy for decentralised domain management, based on a proven hierarchic data structure. On the way of developing a framework where individual sub-domains only have local knowledge of their surroundings, this contribution describes the communication patterns used in ensuring a consistent domain, without the need for expensive global broadcast messages. Furthermore, the routines necessary to deal with adaptive changes in domain topology, due to refinement, coarsening and migration of sub-domains to different computational resources, are discussed in detail.