Welcome to the 7 Questions for Chapel Users series! In this edition, we welcome Dr. Marjan Asgari. As an expert in computational hydrology, Marjan used Chapel to improve hydrological models as part of her Ph.D. at University of Guelph. Read on to learn more about her journey using Chapel!
1. Who are you?
My name is Dr. Marjan Asgari, and I am a Physical Scientist at Natural Resources Canada, Government of Canada. My primary expertise lies in the parallelization of large-scale geospatial tasks and working with extensive geospatial datasets. I regularly work with HPC clusters and enjoy tackling projects that face challenges with execution speed and memory exhaustion, applying “parallelization magic” to optimize them.
“It is important to choose approaches that are not only easy to implement, but also easy for scientists—whose focus is not on computer science—to understand.
”
2. What do you do? What problems are you trying to solve?
My main role involves solving computational and memory-related issues in geoscience, particularly when using machine learning models for large satellite datasets. When dealing with large datasets and complex tasks, the need to transition to novel computational approaches becomes evident. However, it is equally important to choose approaches that are not only easy to implement, but also easy for scientists—whose focus is not on computer science—to understand. So, constantly researching for novel approaches in parallel computing and implementing them for geoscience projects is my daily job.
3. How does Chapel help you with these problems?
I think the most important feature of Chapel is its ease of use. Parallelization is not an easy task to accomplish; the whole concept of how to parallelize your work efficiently is complex enough. Coding this concept in a language that is difficult to use only adds to the challenge and discourages scientists from giving it a try. Chapel makes coding engaging and increases curiosity about how to enhance parallelization in your code. I have only run Chapel on a multi-node cluster with CPU-based compute nodes, but with the increasing popularity of deep learning models in my field, I believe the next step for me will be GPU processing.
“Chapel makes coding engaging and increases curiosity about how to enhance parallelization in your code.
”
4. What initially drew you to Chapel?
Before working with Chapel, I was using parallel computing frameworks like Spark and Hadoop. The challenge of simply setting up a distributed system and scheduling tasks among parallel workers was immense. Then, I started researching other possible ways of doing parallel computing. I came across Chapel, ran a coforall loop, gathered all the results outside the loop, and I thought, “Wow, this is it!”.
5. What are your biggest successes that Chapel has helped achieve?
My Ph.D. dissertation represents a significant success. I was able to develop a method for the optimization of hydrological models using Chapel. There were no publications on the use of PGAS (Partitioned Global Address Space) models in the calibration of hydrological models, and Chapel enabled me to contribute to this field with my work.
6. If you could improve Chapel with a finger snap, what would you do?
I really wish the installation of Chapel on HPC systems were easier and more straightforward. Relying heavily on SSH as the start-up method can be insufficient, as many HPC systems don’t permit SSH between compute nodes. I put in a lot of effort to [note:After Marjan completed this work, we developed a Spack package for Chapel, which should significantly simplify the process of installing Chapel on HPC systems. Sorry we didn’t have it at that time, Marjan!] we had, but unfortunately, it didn’t always work in the end.
Another item on my wish list is that Chapel would support geospatial data. If Chapel could work with data structures [note:Chapel doesn’t natively support NumPy arrays or XArrays. However, Arkouda does! We recommend checking Arkouda out if you are interested in using such technologies interactively at HPC scales.], it would be a game-changer. In the field of Geomatics, there is immense potential for using parallel computing with languages like Chapel, which are not difficult for Python programmers to learn. The lack of spatial data support is a significant limitation in this regard.
7. Anything else you’d like people to know?
As a scientist working in the field of parallel computing, Chapel has amazed me several times. The only thing I hope for is that the items on my wish list in question 6 come true at some point. If that happens, I believe working with Chapel will be a blast every day.
We’d like to thank Marjan Asgari for participating in the 7 Questions for Chapel Users series. If you have any questions for Marjan, or comments, please direct them to the discussion thread for this series on Discourse.
