HomeNews and blogs hub

HPC enters a solid new discipline: ParaFEM

Bookmark this page Bookmarked

HPC enters a solid new discipline: ParaFEM

Author(s)

Lee Margetts

Posted on 13 May 2014

Estimated read time: 4 min
Sections in this article
Share on blog/article:
Twitter LinkedIn

HPC enters a solid new discipline: ParaFEM

Posted by s.hettrick on 13 May 2014 - 9:33am

By Lee Margetts, University of Manchester.

Scientists use a variety of computational techniques in their research, and many disciplines benefit from access to High Performance Computing (HPC) to solve their grand challenge problems. There are established communities with mature HPC codes in physics, earth science, cosmology and more. Computational chemists recently won the Nobel Prize for their efforts, highlighting the benefits of investing in high quality scientific software for HPC. I believe that computational solid mechanics can now join the HPC club.

In computational engineering, the use of regional, national and international HPC facilities is dominated by software for modelling fluids. In a recent NAFEMS World Congress, one presentation reported the use of 200,000 cores to study turbulent flow around aircraft, whilst another on solid mechanics used a multi-core desktop to increase processing speed by four times using only eight cores and two graphics processing units. The flow modelling used the open-source program OpenFOAM, and the solid mechanics example used a well-known commercial application. The juxtaposition of these talks highlighted the difference in capability and the widening gulf between modelling fluids and solids on supercomputers.

With support from the Software Sustainability Institute  I am working to develop computational solid mechanics for HPC. This is important, because it could lead to positive knock-on effects for UK innovation and GDP in key sectors such as Aerospace, Energy and Transport. Modelling solids has applications in many important and topical areas, such as assessing the environmental impact of fracking, ensuring the safety of nuclear reactors and exploring the manufacturing potential of the wonder material graphene.

A few years ago, I initiated the open source ParaFEM project (#ParaFEM), a massively parallel library for finite element analysis that can solve engineering problems with billions of unknowns on tens of thousands of cores. The foundation is an engineering text book, Programming the Finite Element Method, which was first published in the 1980s and is now in its 5th edition. Recent development effort in ParaFEM has concentrated on building a platform for sustainable growth in terms of both usage and developer contributions. The latter is particularly important if ParaFEM is to have functionality on a par with commercial packages.

In the autumn of 2013, my colleague Louise Lever and I applied to the Institute’s open call for collaborative projects. We are very pleased that the Institute agreed to help us in three areas and assigned their consultants John Robinson and Steve Crouch to the project.

First in our wish list was a quick demonstration installer for desktop Windows platforms. This may seem to be counter-intuitive for software that runs on a supercomputer, but the rationale is that potential users and contributors surfing the web may spend only a minute or two to assess an open-source project. The quick demonstrator allows potential users to interact with the software on their own computer. If they like what they see, they may be more likely to invest in the time required to download the full package and evaluate it on an HPC system.

We also asked the Institute to help implement a post-installation test process - a basic but important feature missing from the current distribution. ParaFEM is a library with example programs or “Mini Apps”. The test process will simply check that the programs have been built correctly.

Finally, the main aim of our work with the Institute is to find out what we are doing right (or wrong) in managing our open-source project. We view the Institute as an essential partner in helping us achieve our ultimate goal, to turn ParaFEM into the open source solid mechanics equivalent of OpenFOAM.

Share on blog/article:
Twitter LinkedIn