Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GPU AND CLOUD COMPUTING FOR ENGINEERING
Edited by: P. Iványi and B.H.V. Topping
Automated parallel and distributed computation of finite element equations with Python descriptors and DASK framework
Department of Civil Engineering, Faculty of Civil Engineering, Istanbul Technical University, Istanbul, Turkey
M. Yilmaz, "Automated parallel and distributed computation of finite element equations with Python descriptors and DASK framework", in P. Iványi, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Parallel, Distributed, GPU and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 4, 2019. doi:10.4203/ccp.112.4
Keywords: finite element, distributed computing, parallel computing, Python, descriptor protocol, DASK.
In this study, we present a domain specific modelling language (DSML) in Python for rapid prototyping of Finite Element (FE) theory. In the DSML, we provided the developers with generic programming support for the construction and solution of discretization schemes, in the context of partial differential equations. We take advantage of Pythonís descriptor protocol and the DASK framework (a flexible library for parallel/distributed computing in Python that is optimized for interactive computational workloads) to facilitate potentially tedious parallel/ distributed FE programming tasks in a self-explanatory syntax. For that, we offered several automated utilities in the form of individual descriptive-objects, namely descriptors, with additional support for lazy-evaluation. We then allowed the developer to inject these descriptors as natural dependencies into their custom classes by taking advantage of Pythonís descriptor protocol. By using concrete examples, we demonstrate that adopting these descriptors to use the DASK-scheduler on the background results in a concise, efficient and customizable parallel/distributed code base for FE analysis.
purchase the full-text of this paper (price £22)