A distributed computation library for common modeling tasks in magnetic confinement fusion

FusionSC is a Python and C++ library designed to give researchers tools for common data processing tasks, such as field line tracing, heat load estimation, magnetic surface reconstruction, and mesh geometry processing. FusionSC contains optimized C++ kernels for such operations, which can be used either locally or via remote procedure calls. Complex abstract machine descriptions are supported, including processing trees and named references (which are automatically resolved through lookups in local and remote databases). A distributed data protocol transfers data from local and remote sources to required nodes, using hashes to avoid duplicate transfer and storage. Cap’n’proto, a fast zero-copy serialization scheme and RPC protocol, serves as a basis for both storage and communication, supporting usage patterns from small local installs with local component databases to scalable, load-balanced cloud deployments with dedicated compute and storage nodes. This manuscript discusses the overarching design goals of FusionSC, introduces the distributed data and processing architecture, summarizes the most relevant physics capabilities, and illustrates how the various subsystems interact to enable complex tasks such as heat load computations.