Thinking about stars … in the light of supercomputers
With a bag of disposable earplugs in one hand, I was about to enter the noisy large hall hosting two of the computers we were using at the time for some star simulations.
This article is written by Maria Guadalupe Barrios Sazo
It was a fresh day in Berkeley, California and the floors under these large machines were carefully designed to survive strong earthquakes. The computers were covered by cases with beautiful prints. One of them was named Cori, after the biochemist who earn the Nobel Prize in Psychology and Medicine in 1947.
The thing about me is that I find computers exciting, thinking about their history, the charm in the variety of devices and the ways we interact with them. I am not by any means a hardware guru or an algorithms prodigy, but there is a feeling of belonging when I think about computational science and high-performance computing. Many people start in astrophysics because of the beauty in trying to understand the mysteries of our universe. Myself (and maybe others?) because we like the beauty of using computers in challenging ways plus cool mathematics when applied to astrophysics.
(Don’t get me wrong, I grew up wanting to be an astronaut)
For my PhD research, I was lucky to join the computational astrophysics group at my university and then became properly introduced to super computers, some of the applications in astrophysics, and their challenges.
What do we mean by supercomputers?
What people usually refer to as a super computer is a machine that performs a very very large number of operations/calculations per second and we measure those in a unit called FLOPS. Since the nineties, twice a year a list with the top 500 supercomputers in the world is released. To rank the computers there is a common challenge of matrix operations that is run on each system. Also, more recently they have an extra category to reward the greenest ones. In a way, you can picture many of them as hundreds of computers connected together. But their components and how they are interconnected can be quite diverse.
Some of the applications in astrophysics
We use powerful machines in astrophysics in many sub areas, ranging from simulations to data acquisition and data analysis. For instance, during my PhD I worked on simulations of white dwarf stars merging. In the techniques we used the stars are modelled like fluids (gases). The physical processes, such as gravity and magnetic fields, are described by equations that require a lot of computation power to be solved.
Now I work with a group that simulates multiple aspects of the sun. In both cases, using supercomputers allows us to achieve better results by including more physics, volume, and finer resolutions.
Challenges of using supercomputers
To have these immense computing facilities operational and useful for the users there is a lot of dedicated people working since early planning until the system is decommissioned. The space holding this computer has to be sturdy to hold their weight and spacious to allocate all the components. The cooling system has to be reliable to handle all the operations happening when the system is fully used. The amount of energy is large which leads to environmental concerns and addressing them has become a big priority.
How does it look like to work on supercomputers?
The truth is that although we get to use these cool machines, many of us spend a lot of time in prototyping, reading applied math papers, developing, deploying, and testing the codes we and our fellow astrophysicists use. For this process, our codes ideally run on our personal computers with much smaller and simpler models. Then we move to larger setups on the university/group clusters and get an idea of the efficiency in our codes. From there, we can make a case that our codes would benefit and fully exploit the resources in a supercomputer. When using the super computers, we connect remotely from our personal machines (a lot of security on this step), we set up the environment compatible with our problem (for instance the compilers we would like to use), we build/ compile our code and then line up our simulation in a queue to wait for the resources we are requesting.
I hope in the words above I projected that the use of super computers is a constantly evolving field with tons of wonderful people from different disciplines and interests that converge to advance science and technology -- including our knowledge of the universe.
