In this edition, we profile Dr Arvind Iyer, Postdoctoral Research Student, Imperial College London, who is focusing on high order methods in CFD.
I studied for a Bachelor of Engineering degree in Mechanical Engineering at Mumbai University between 1999 - 2003. Following this, I completed my Masters in Aerospace Engineering and a PhD at the Indian Institute of Science in 2014, focusing on Large Eddy Simulations of the supersonic mixing layer using OpenFOAM with applications to Scramjet Engines. In December 2014, I came to London and I was able to secure a position as a Postdoctoral Researcher at Imperial College London (ICL) in the Department of Aeronautics. Here, I am currently working of the development of PyFR software, an open-source Python based framework for solving advection-diffusion type problems on streaming architectures using the Flux Reconstruction approach of Huynh. Besides, we apply PyFR in the study of compressible flow over cavities with applications to noise reduction, flows over low pressure turbine blades cascade with applications to improving airplane engine efficiency and turbulent flows in rectangular channels with applications to improving turbulence models.
I have been working in the field of Computation Fluid Dynamics (CFD) for more than 12 years, and throughout my area of interest has been in developing and using technologies closely aligned with the requirements of the industry. My career aspirations are a continuation of the same interest, where I want to focus on development and usage of niche technologies that are closely aligned to the requirement of the industry and will deliver innovative benefits.
In the past decade, there have been a lot of innovation and development of massively parallel accelerators with direct application to High Performance Computing (HPC). As compared to a pure CPU implementation, computations need to be set up in a slightly differently way to make an optimal use of these accelerators. However, its often difficult to cast traditional CFD algorithms, like the finite volume method, to make use full use of the accelerators. High order methods, like the flux reconstruction method, in contrast, are much more suitable for speedup using these accelerators. For example, using PyFR, we have been able to demonstrate an efficiency as high as 58% of the theoretical peak simultaneously attaining a sustained throughput of more than 13 Petaflops. In contrast, finite volume based approaches seldom exhibit an efficiency greater than 10% of the theoretical peak. I do believe that there is a scope for the industry to adapt to emerging technologies like the Graphical Processing Unit (GPU) Computing and algorithms like high-order methods, and I wish to be a part of this adaptation process.
My most notable achievement has to be where I was part of the team at ICL who were shortlisted as a finalist for the Gordon Bell Prize (HPC Applications category) and SC16’s Best Paper Award. Our submission covered the topic of achieving accurate simulation of unsteady turbulent flow, which is critical for improved design of ‘greener’ aircraft that are quieter and more fuel-efficient, demonstrated by application of PyFR, a Python based computational fluid dynamics solver, to petascale simulation of such flow problems. It was a great achievement and an accomplishment to be shortlisted in a worldwide competition. I am also currently a principal investigator in a project related to turbulent channel flows studying one of the fundamental aspects of turbulence, which is very interesting.
I have met a wide range of people throughout education and industry, and I would have to say my teachers, my PhD and Masters supervisors, and my supervisor at ICL have been an inspiration to me.