Pairwise verlet lists: combining cell lists and verlet lists to improve memory locality and parallelism

Area of Science:

• Computational physics
• Molecular dynamics simulations

Background:

• Traditional global Verlet lists are inefficient on modern parallel multicore architectures.
• Existing methods struggle to scale effectively with increasing core counts.

Purpose of the Study:

• Introduce pairwise Verlet lists for particle-based simulations.
• Evaluate their performance and scalability on shared-memory systems.

Main Methods:

• Developed local Verlet lists containing only interacting particle pairs between neighboring cells.
• Compared pairwise Verlet lists against traditional global Verlet lists.

Main Results:

• Pairwise Verlet lists demonstrate superior efficiency on single processors.
• Significant improvements in scalability observed on multiple shared-memory cores.
• Enhanced performance on single cores also benefits distributed-memory simulations.

Conclusions:

• Pairwise Verlet lists offer a more efficient and scalable alternative to global Verlet lists.
• This method is well-suited for modern shared-memory parallel architectures.
• The approach shows promise for distributed-memory simulations as well.