Towards multicluster computations with Julia
Francisco Heron de Carvalho, Tiago Carneiro · 2024
The ability to aggregate the computational resources of multiple clusters is useful for solving large problems that can benefit from multicluster platforms. In this context, this paper introduces an extension to Julia that enables multilevel parallelism and allows users to exploit use cases of multicluster computation. The proposal is evaluated through a proof-of-concept case study based on the multizone version of the NAS Parallel Benchmarks (NPB-MZ), focusing on evaluating inter-cluster communication and load-balancing overheads.