Introduction
Every year, in collaboration with RCC Director Professor David Abramson, RCC and The University of Queensland, four secondary students are selected to visit the Supercomputing Conference in the USA.
Our four Queensland Academy for Science, Mathematics and Technology (QASMT) students this year, Magnus, Murphy, Eton and Olivia, were selected to attend based on their involvement in computer science and information technology in a global society (ITGS) courses. In these courses, students had the opportunity to explore the application of high-performance computing (HPC) in the fields of science, astronomy, machine learning and a plethora of other fields.
This curriculum link gave the students a much clearer idea of what to expect from SC19, and the greater industry of HPC.
Below are the students’ reflections on their experience at SC19.
SC19 Reflection
SC19 was held in Denver, Colorado from 17 to 22 November. This year we had the privilege to represent QASMT and Australia’s involvement in HPC at this annual supercomputing conference.
SC19 was an incredible experience that showcased amazing advancements in technology and alluded to the future of supercomputers and computing.
Throughout the conference, we engaged with key speakers, people and businesses that all were passionate in using HPC to benefit themselves and advance humanity.
However, the conference was not only an opportunity to marvel at new advancements and gasp at water-cooling PCs, it was also an experience to learn and discover new and interesting things.
It truly was a pleasure and a privilege to be selected and attend such a prestigious event in the IT world. In the true spirit of SC19, we can all agree that ‘HPC is now’.
Magnus: D-Wave Systems
D-Wave Systems is currently the leader in commercial quantum computing technology, and at SC19, the Canadian company hosted a seminar “Introduction to Quantum Computing & Programming”.
Quantum computing is my desired field of study at university and a possible career path, so I was looking forward to the talk.
The talk opened with Rene Copeland of D-Wave presenting a general introduction to the company’s quantum computer and its current applications, most notably the optimisation of bus routes and traffic flow by Volkswagen Group.
Rene went on to talk about the basics of their QPUs (Quantum Processing Units), which are based on niobium metal and currently have 2,048 qubits using the annealing method.
What particularly interested me was the new Pegasus Quantum Computer that Rene touched on, and later D-Wave physicist Edward Dahl discussed in detail.
The talk also touched on the methods of programming and the full stack for program development for quantum computers, and how programmers can best take advantage of quantum computing.
The new generation of D-Wave quantum computers, codenamed Pegasus, are “drastically different from the previous generation”. They have 5,000 qubits, which is 2.5x the last generation, however what was most interesting was the inter-qubit connections in the form of Josephson Junctions (JJ’s), which they were almost able to triple.
This means that qubits can influence each other easier and can then lend themselves to larger and more challenging problems—think multithreaded programs in the classical world.
They were able to achieve this by placing the qubits in a special polygonal lattice structure that allowed for the greatest interconnects between qubits (see the image provided below).
I would think that tasks such as this—finding specific mathematical structures that allow for the most possible connections (JJ’s) and the greatest possible number of corners (qubits) in a fixed space and the closest possible length connectors—could be aided by HPCs or even QPUs in the future.
Another aspect of the new QPUs that really interested me was the addition of a capacitor into the quantum circuitry on the QPU, which added extra terms to the Hamiltonians that was theoretically shown to increase the “jumps” between energy states and proven experimentally to overall increase single-qubit performance.
The most important part of this talk, for me, was the experience of being immersed into a specific part of a field I was very interested in, and the exposure to current technologies and what is actually happening now from the company itself, instead of what the media provides.
Murphy: Keynote and Invited Talks
Personally, the keynote talk excited me as it covered planetary exploration and HPC involvement.
Prof. Steve Squyres, principal investigator of the Mars Exploration Rover Mission, detailed his experience with the Spirit and Opportunity rover programs. He highlighted how the rover missions have opened our eyes towards Mars, its geological and seismic history, as well as why it is like it is today.
HPC becomes relevant when computational fluid dynamics are required with the rover capsule entering the Martian atmosphere at Mach 27 (9261m s-1) and the supersonic parachute deployed at Mach 2 (686m s-1), thus some serious calculations are needed to ensure the rover arrives safely.
At this velocity, their supersonic parachute requires meticulous design, thus supercomputing is utilised for computational fluid dynamic calculations that would otherwise have been nearly impossible to complete in a reasonable time frame.
Prof. Squyres explained:
- challenges and roadblocks that were dealt with
- equipment and instruments, and
- some accidental discoveries and lucky saves.
The highlights for me were (in no particular order):
- Due to the compressed time frames before launch, the program for driving and pretty much everything that was needed once on the surface of Mars was written after launch, in the seven months of space flight.
- They named a dune on Mars “Purgatory Dune” because of all the problems it caused them.
- When Spirit’s front right wheel broke and they decided to drive the rover backwards dragging the dead wheel through the soil, a white substance was spotted in the trench dug. It was analysed to be about 91% pure silica, naming it justly “Silica Valley”. This discovery was astonishing because the only way that a silica deposit that concentrated would be formed, is through processes relying on the presence of water.
- Coming back to supercomputing, planetary exploration requires HPC for many calculation-heavy processes, from hypersonic flight and the parachutes (computational fluid dynamics) to trajectory design and propulsion.
Overall, there is so much more I could have written about this talk.
Additionally, at SC19, were the invited talks; I especially enjoyed the astrophysics-orientated ones. I will describe two briefly.
Prof. Orsola De Marco talked about astrophysics and the use and limits of HPC. She outlined what are considered the “three big problems” of astrophysics, being:
- planet formation (having no discrepancy between simulations)
- stellar collisions
- evolution of the universe. Further explaining HPCs’ large role in the field and in learning more about these phenomena.
Plus, Prof. Anders Ynnerman’s presentation on OpenSpace and how it was created. OpenSpace is an open source program that visualises astronomical data sets amazingly as well as generating them in quite a photorealistic fashion.
There is so much I could say about the SC19 program but since it is open source, I encourage you to go look.
Eton: Hackathon
While at SC19, we were invited by the organisers of the Cloud HPC Hackathon to watch the final presentation and judging of the programs submitted by participants.
The participants all had varying challenges they needed to complete in new and innovative ways.
These challenges led to the construction of a Python module that allows scientists to fetch and visualise pre-existing data housed in chord diagrams using Jupyter Notebook, a user interface to execute, manage and monitor jobs on CloudyCluster via both JupyterLab and html/js UI, a Twilio chatbot that allows users to submit and manage HPC jobs and check status via text messages, and a parallel malware analysis gateway that provided disassembled data from submitted executable malware samples.
These programs not only had to be an original idea constructed from scratch, but they also had to be done within 44 hours.
Therefore, fighting time restrictions and other technical issues, including having to learn entirely new programming languages within this time frame for some groups, the participants were able to produce exceptional, but by no means flawless, programs.
While the majority of the programs could be significantly improved, what they were able to build within the time was very impressive.
After seeing what the participants were able to construct in 44 hours with limited resources, technological issues and with great team spirit (though teams were in competition with each other they treated it as a collaborative environment by helping troubleshoot issues and optimise solutions), I left feeling inspired to participate in a hackathon back in Australia.
This event has really shown me the extent of what a team can accomplish in 44 hours when you have the right skills, attitude and mindset.
Olivia: Exhibits
At SC19, I particularly enjoyed the exhibits. This year there were 404 exhibits—ranging from major technology companies, such as Microsoft and Intel, to current leaders in HPC, such as Google, to universities from Indiana and Texas, hardware suppliers, such as Gigabyte, Intel and NVIDIA, as well as companies using applications of HPC for scientific research, such as NASA and even our own NCI Australia and Perth-based Pawsey Supercomputing Centre.
I appreciated the opportunity to talk to these companies and their employees to see where a future in IT and HPC could take me in the future.
It was interesting to learn and discuss the applications and intentions of HPC within these businesses and the importance of HPC to society.
Thanks to Dennis Sullivan from Lenovo for enlightening me on the status of HPC being used in software development for purposes such as artificial intelligence and machine learning. He directed me to a SC19 workshop by Bing Xie who presented “Applying Machine Learning to Understand the Write Performance of Large-scale Parallel Filesystems.”
The workshop was a bit technical and advanced for me, but what I did learn was in input/output performance prediction could be more efficient in scientific computing by using HPC. For example, the workshop identified what factors most affected predicted input/output performance in order for more accurate predictions resulting in more precise runtime estimates.
As I did not fully understand all components of computer hardware, my mission in coming to SC19 was to gain greater understanding of hardware and processing inside computers.
After some intense discussions at Intel, I learnt about their new programming model called 'One API'. This is big news for the industry, as in simple terms, the One API model will allow developers to reuse the same code when the underlying hardware changes. This new model will unify code across multiple hardware targets like Nvidia and Intel GPUs.
In addition to this new announcement, it was unveiled that Intel and Cray are building a $500 million ‘exascale’ supercomputer for Argonne National Lab. This supercomputer will have nodes with two ‘Sapphire Rapids’ CPUs and 6 ‘Ponte Vecchio’ GPUs with a unified memory architecture across the CPU and GPU.
SC19 was truly the place to see all the new and innovative advancements in the IT world and I can definitely say I learnt more about computer hardware throughout this event.
Aside from all the new announcements unveiled at the exhibits, I know everyone enjoyed unleashing our inner nerd playing with lightsabres and virtual reality headsets, with one lucky person in our group even winning Apple’s new product, AirPods Pro.
Conclusion
The students have brought home a wealth of knowledge from SC19 that they are planning on utilising within the school.
We have invested in a collection of Raspberry Pis and plan to build our own small cluster computer, based on what these students have learned.
They will also be providing leadership to students next year who study our HPC units by sharing their experiences and resources they acquired from SC19.
We want to express our deepest thanks to Professor Abramson, RCC and The University of Queensland for providing such an enriching experience for our students and our school community.
Introduction and conclusion by QASMT Computer Science, ITGS, Design and Robotics teacher Simon Freeman.
