Computational Cosmology and Astrophysics on Adaptive Meshes using Charm++

TL;DR

This paper introduces Enzo-P, a scalable astrophysics simulation framework built on the Cello adaptive mesh refinement software, leveraging Charm++ for high scalability on petascale and future exascale supercomputers.

Contribution

It presents a novel scalable framework for astrophysics simulations using adaptive meshes, based on Charm++ and designed for extreme-scale computing environments.

Findings

Achieved scalable performance on NSF Blue Waters supercomputer.

Demonstrated the effectiveness of Charm++ in adaptive mesh refinement applications.

Outlined plans for adapting Enzo-P to exascale heterogeneous platforms.

Abstract

Astrophysical and cosmological phenomena involve a large variety of physical processes, and can encompass an enormous range of scales. To effectively investigate these phenomena computationally, applications must similarly support modeling these phenomena on enormous ranges of scales; furthermore, they must do so efficiently on high-performance computing platforms of ever-increasing parallelism and complexity. We describe Enzo-P, a Petascale redesign of the ENZO adaptive mesh refinement astrophysics and cosmology application, along with Cello, a reusable and scalable adaptive mesh refinement software framework, on which Enzo-P is based. Cello's scalability is enabled by the Charm++ Parallel Programming System, whose data-driven asynchronous execution model is ideal for taking advantage of the available but irregular parallelism in adaptive mesh refinement-based applications. We present scaling results on the NSF Blue Waters supercomputer, and outline our future plans to bring Enzo-P to the Exascale Era by targeting highly-heterogeneous accelerator-based platforms.