Table of Contents

This is the first in a new series of articles we’re launching in which we’ll be asking Chapel users a series of questions about their work and experiences with the language. In doing so, our goal is to shine a light on ways in which Chapel is being used, and to hear from users in their own words. If you are using Chapel and would like to be considered for a future installment, or you know of someone who is and ought to be, please let us know!

For our inaugural edition of this series, we turned to Éric Laurendeau, whose team has developed one of Chapel’s most ambitious flagship applications, CHAMPS. We could say more to introduce Éric and CHAMPS, but he does a perfect job of it himself, so let’s dive right in!

1. Who are you?

My name is Éric Laurendeau, professor at Polytechnique Montréal. I’m fortunate to have earned my Bachelor’s in Canada (McGill), Master’s in France (ISAE-SUPAERO), and Ph.D. in the USA (U. Washington). Why? These are the top 3 worldwide aerospace hubs. Montreal is home to Bombardier (business jets), CAE (flight simulators), Airbus Canada (Airbus 220), Pratt & Whitney (engines), Toulouse is Airbus’s final assembly line and Seattle is Boeing’s place. So I have had a chance to see various approaches to aircraft design and manufacturing.

From 1996–2011, I worked in the advanced aerodynamics department of Bombardier during the ‘golden years’ of Computational Fluid Dynamics (CFD). Indeed, major scientific breakthroughs combining numerical algorithms and fluid dynamics during the 1980’s, combined with advances in computer power, gave rise to this ‘3rd discipline’, complementing experimental and theoretical fluid dynamics.

The design of aircraft aerodynamics is now done almost exclusively using CFD, and the final shapes are validated in a wind tunnel and then flight-tested for certification purposes. Since aerodynamics is very closely linked to advances in state-of-the-art research, I took the challenge of becoming a professor in 2011 to concentrate on the development of novel computational aerodynamics methods while teaching at the undergraduate and graduate levels. I typically direct the work of some 15 Masters and Doctorate students along with one or two post-doctoral fellows funded through fundamental grants from Canada and industrial grants, complemented with federal and provincial governmental grants.

2. What do you do? What problems are you trying to solve?

The aerospace industry is more than ever trying to reduce its environmental footprint, whether CO2 or noise emissions. To achieve its own goal of net zero emissions by 2050, novel aircraft configurations like NASA’s Transonic Truss-Braced Wing or Blended Wing Body concepts are examined (you can look at these on the internet!). Moreover, since air traffic is ever-increasing, and atmospheric conditions changing due to global warming, aircraft safety must be addressed.

“

The aerospace industry is more than ever trying to reduce its environmental footprint, whether CO2 or noise emissions.

”

My laboratory therefore examines:

One thing all these models share is the need for supercomputers. Indeed, several of these models solve for some 10 million to 1 billion unknowns! Thus, applied mathematics and computer science know-how must be mastered to enable these aerodynamic models to be used by industry to solve today’s challenges in aviation.

3. How does Chapel help you with these problems?

To solve for these incredibly large systems of equations, supercomputers must be used. In the past century (until year 2000) these were developed using shared memory systems (e.g., several computer chips using one brain), and chip synchronization was performed using the OpenMP protocol. The early 2000’s gave rise to distributed memory systems (e.g., several computer chips using several interconnected brains) using the MPI protocol. These protocols add several layers of complexity to already very complex software. The result is so top-heavy that it can sometimes lead to software being too complex to continue further developments (like simply reaching the limit of highway traffic, where every car literally is stuck on the freeway). One alternative to these low-level languages is Python, a very easy-to-program, interpreted language which is very inefficient for scientific computing.

“

Chapel removes the layers of complexity for code writing, allowing seamless code development while maintaining computational performance. It is ideal to develop very large and complex computational models.

”

Chapel was designed exactly to address this gap in a very elegant and efficient way. It removes the layers of complexity for code writing, allowing seamless code development while maintaining computational performance. It is ideal to develop very large and complex computational models. Another great feature is its simplicity. Thus, when students or staff (e.g., post-docs, researchers) come and go in the software, the training or time-to-understand is very significantly reduced while reducing the chances of errors or bugs popping up.

Chapel is such an integral part of our laboratory’s success that our main aerodynamic solver is named CHAMPS, standing for Chapel Multi Physics Software! We run CHAMPS on our desktops and laptops. This provides flexibility when students perform out-of-country or outside the laboratory exchanges, simply bringing their laptops with them. Once simple models are made, very complex ones are launched on the supercomputers of the Digital Research Alliance of Canada which are free to use for Canadian researchers (albeit with a cap on the number of core-years used).

4. What initially drew you to Chapel?

While there exist several commercial aerodynamic software packages, and even open-source ones (industry typically uses proprietary software), I needed to develop in my lab new software to solve 3D aerodynamic problems to cater for my laboratory’s long-term research objectives. All aerodynamic software is either using Fortran (yes, you read correctly!) or C/C++, while all use the MPI protocol. Having developed in my laboratory a 2D code in C with OpenMP instructions that had grown so large that it became too difficult to maintain, my students stumbled on Chapel by chance. They wrote emails to one of the team developers, Dr. Brad Chamberlain, to ask naïve questions and were surprised to receive answers to every one of these and with a very rapid response time! They thus took the bold step to write a prototype code to see its performance, which ultimately was as good as advertised! They then proceeded to write a more comprehensive version that has kept growing ever since with new models/capabilities.

“

The use of Chapel worked as intended: the code maintenance is very much reduced, and its readability is astonishing. This enables undergraduate students to contribute to its development, something almost impossible to think of when using very complex software.

”

5. What are your biggest successes that Chapel has helped achieve?

We managed to very rapidly develop a state-of-the-art 3D aerodynamic solver that runs on shared memory and distributed memory systems. Furthermore, the solver is written in a very compact form, with some 10x reduction in the number of instruction lines compared to similar software written using Fortran, C, or C++ with OpenMP/MPI overhead. The use of Chapel worked as intended: the code maintenance is very much reduced, and its readability is astonishing. This enables undergraduate students to contribute to its development, something almost impossible to think of when using very complex software. Students enjoy working on a new technology while contributing to the community’s efforts to address the challenges of solving very complex mathematical models using simple programming techniques. This enables researchers to focus on novel contributions rather than spend energy on writing lines of codes with associated debugging time.

Aircraft Aerodynamic modeling using Chapel on High-Performance Computers

Aircraft Aerodynamic modeling using Chapel on High-Performance Computers

CHAMPS enabled significant scientific contributions. For instance, it helps develop new models that are then integrated into the R&D stream of industrial companies, such as aero-icing capabilities or state-of-the art global stability analysis of important aerodynamic phenomena. It is also used in the many American Institute for Aeronautics and Astronautics (AIAA) workshops intended to evaluate the state-of-the-art in high-lift (take-off/landing) or drag (cruise flight) simulations. CHAMPS thus stands on par with similar software developed by national research centers such as NASA (USA), ONERA (France), DLR (Germany), JAXA (Japan), etc., as well as commercial vendors (Ansys, STAR-CCM+, etc.). Of these, it is the only one not using OpenMP/MPI protocols and the only one using Chapel! These activities showcase the feasibility of a new parallel paradigm for the aerospace community. Indeed, High Performance Computing is explicitly listed as a mandatory ingredient in NASA’s vision for the future (you can read this report: CFD Vision 2030 Study: A Path to Revolutionary Computational Aerosciences, available on the web). Other contributions are in computational aeroelasticity, laminar-turbulent transition of airflow, and many others: you can simply see the list of publications on my University website.

“

CHAMPS thus stands on par with similar software developed by national research centers such as NASA (USA), ONERA (France), DLR (Germany), JAXA (Japan), etc., as well as commercial vendors (Ansys, STAR-CCM+, etc.). Of these, it is the only one not using OpenMP/MPI protocols...

”

CHAMPS has significantly contributed to reach many of the scientific objectives in my Canada Research Chair award (2017–2024), for which I am awaiting its renewal decision for 2025–2031! It helps attract talent, who is always attuned to novelty, and has attracted industrial contracts. CHAMPS’s success has enabled growth beyond my laboratory: it is used at Université de Strasbourg (France) by Prof. Y. Hoarau to examine novel Immersed Boundary Methods to tackle the geometrical complexities associated with typical aircraft designs, as well as by Prof. Paoli at Polytechnique Montreal to examine aircraft contrails (the white twin-streak clouds behind aircraft at high altitudes) that are the most important contributors to aircraft emissions. CHAMPS is currently being examined for GPU computing, again via the capabilities of the Chapel language and to develop very accurate yet very expensive Hybrid Reynolds-Averaged Navier-Stokes models, which are time-dependent (increasing the number of aircraft solutions by a factor of 1000) and mesh discretization dependent (increasing the number of discrete space solutions to some 200 Million cells, making for some 1.4 Billion unknowns!).

6. If you could improve Chapel with a finger snap, what would you do?

The comments received from my students for improving Chapel are:

7. Anything else you’d like people to know?

Students who leave the laboratory after graduation or to pursue further studies in another laboratory often tell me they miss Chapel!

Thanks very much to Éric for taking the time to kick off this interview series with us! If you’d like to hear more about the CHAMPS team’s work, including an introduction to the use of HPC in computational aerodynamics, be sure to check out the video or slides from Éric’s excellent CHIUW 2021 keynote, HPC Lessons from 30 Years of Practice in CFD Towards Aircraft Design and Analysis.

And/Or, for those attending SC24, Éric will be the Distinguished Speaker at PAW-ATM 2024—the 7th Annual Parallel Applications Workshop, Alternatives to MPI+X—and we highly recommend catching his talk, A Case Study for using Chapel within the Global Aerospace Industry.

Finally, if you have any other questions for Éric, or comments on this series, please direct them to the 7 Questions for Chapel Users thread on Discourse.