Rupak Biswas, Ph.D., director of NASA’s exploration technology directorate at the Ames Research Centre in Silicon Valley recently visited the European Investment Bank. His visit followed his attendance at the ISC High Performance conference in Frankfurt where EIB’s Innovation Finance Advisory (IFA) launched its study on financing supercomputing in Europe. That made it a good moment for Rupak to share his perspective on the topic in conversation with IFA head Shiva Dustdar.
Shiva Dustdar: Our study shows that several European countries have regional or national funding plans for supercomputers. However, these initiatives often lack coordination and tend to be fragmented, resulting in underinvestment for strategic High Performance Computing (HPC) infrastructure in Europe. How did NASA realize that supercomputing is worth investing in, and what is the situation in general in the USA?
Rupak Biswas: NASA realized a long time ago that supercomputing is important for us to efficiently do the things that we do. Because the other option is to do physical experiments. But physical experiments are very difficult, they can be extremely dangerous, they can be expensive, they can take a lot of time, and certain physical experiments you just can’t do at all. If you want to check what is the effect of ocean temperatures going up by 2 degrees, you are not going to raise the temperature of the ocean and see what happens. So simulations are a big part of supercomputing. But, with any simulation, your model and your assumptions will determine what your results are.
Certain communities have already understood the benefits of supercomputing. For example, Earth sciences, computational chemistry, fluid dynamics, or aeronautics - many in those communities already know the value of supercomputing.
Many of the big organizations also know the value of supercomputing. Airbus, Boeing, Lockheed Martin, companies like these know that supercomputing can give you a competitive advantage.
And then, of course, there are certain countries like the US and China who have realized that being leaders in supercomputing helps create jobs and benefits the economy, in addition to bringing bragging rights.
It’s the missing middle, we call them sometimes: companies or outfits that are midsize, that don’t understand how to effectively leverage it yet.
The Missing Middle
Shiva Dustdar: These trends are also emerging in Europe. Demand for HPC capabilities is rapidly increasing among large corporates in key sectors of the European economy, such as aerospace, automotive, energy, manufacturing and financial services, while Europe’s more ‘traditional’ SMEs are lagging behind. We see that HPC intermediaries can play an important role in closing the gap. What is your view on this?
Rupak Biswas: The first thing that needs to happen [to bring in this ‘Missing Middle’] is that we need to show these organizations the value of supercomputing by transforming that evidence into economic measures.
Many of the traditional supercomputing people like me will come and say: “Oh, this is so many petaflops, and I can simulate the oceans now in 90 minutes instead of 2 days!
And many people who are not in that field will just roll their eyes. “Okay, so what does that mean?” they will ask. Most people have no idea what a petaflop is, and it is not even clear to them whether a petaflop is a good thing, or not.
Now if you go and say: “If you do this, you will improve your revenue by 2 percent.” – now we’re talking.
I heard that this is what happened with Formula 1 racing cars, for example. They didn’t understand about supercomputing per se. But when they realized that by doing high fidelity modeling and simulation they can cut the lap time around the track by 2 seconds – wow! Suddenly they were interested. Because that translates to revenue, to profitability, to competitive advantage.
Once you understand the potential benefits of high performance computing, you face the next question: How big of a computer do you need to get?
Supercomputing is an undefined term. “Super” by itself has no intrinsic meaning because it’s a relative term – just like if you say something is “big” or “small” it has no meaning unless you say what the baseline is.
Shiva Dustdar: There’s a lot of talk around the Top500 list, and how China has gained positions and relevance in the last few years. How important is it to lead the Top500 list in your view?
Rupak Biswas: Supercomputing people talk about the Top 500 – about how the US and China have a lot of computers in the Top 500, and how Europe doesn’t have that many systems. But we need to look beyond that.
The Top 500 is just based on a number. You run a piece of code, you get a number and a ranking. That doesn’t mean a whole lot. It all depends on what you are using it for and what is the impact.
If you have a lot of money, you can get large systems. Intel, Lenovo, AMD, ARM will sell you systems. But using those systems to actually benefit a field, have an impact, is the bigger challenge. How do you actually use the supercomputer to do something useful?
Shiva Dustdar: Many companies lack technical knowledge about HPC and are therefore finding it hard to make use of HPC services. Without support from experts and a good understanding of the exact business case for the use of HPC applications, companies, in particular SMEs, frequently do not realize the possible economic gains that can be derived from the use of HPC services. Our study shows that advisory services offered by an HPC intermediary can play a critical role in connecting potential users of HPC services and HPC centres. These intermediaries employ trained workers, who can translate users’ needs into HPC solutions and applications. What are the key challenges to increasing demand for HPC services among companies?
Rupak Biswas: It’s just like if you have a lot of money you can buy a lot of gadgets for your house. But then you never use them, because they’re so complicated and they just lie there. You can only tell your friends: “Hey, yeah, I got this thing.”
For example, going back, think about what happened with VCRs. Someone said: “You know, you don’t have to watch live TV anymore, you can get a VCR and you can record it and watch it at your convenience.” Ok, now you’re aware, and then you need the money to go and buy the VCR. So let’s say you make some money and go and buy the VCR and bring it home. But you still don’t know how to use it, because it’s too complicated.
So all three steps are needed. First I need to be aware that a VCR is needed, then I need the funding to get the VCR. And then I need someone to help me to understand how to use it. Otherwise it doesn’t make any sense.
In the HPC context, that requires a trained workforce. You need to bring together domain experts, physicists, mathematicians, chemists who know their field and then you have the computer scientists and computational gurus who know about computers, to actually make an impact. These are people who can bridge that gap between the domain problems that you are trying to solve – whether that is in the financial markets, in pharmaceuticals, whether you’re trying to build self-driving cars, or trying to put in embedded software in drones – and how these problems get translated into computing tasks that you can solve quickly and efficiently. I think that is the next piece of the puzzle, and that’s where China is catching up.
Shiva Dustdar: The issue of a trained workforce is particularly true for smaller companies in Europe. They can often not afford to hire employees working full time on HPC, and this can be a major obstacle to the adoption of this technology. That is why the role of HPC intermediaries should be strengthened. Our report also indicated many banks and investors did not yet fully understand the commercial benefits of HPC to their clients’ financial performance and hence would often not provide financing to help them pay for such services or access to HPC. Many would not have the right financial products on offer – here the EIB Group can also play an important role in offering risk sharing solutions to the financial community.
How super is super enough?
Rupak Biswas: If I go and tell my bosses that I need X dollars to buy a supercomputer, they will ask: “Well, what are the requirements?” That’s a really tricky question.
If I go to my scientists and engineers, I ask them for requirements, they’ll say: “Yeah, I need 10 billion grid points to mesh up this airplane in order to do a proper computer simulation.”
Of course they will.
If you were trying to satisfy the ideals of the scientists and engineers, you could spend a billion dollars on supercomputing, and it still wouldn’t be enough.
Suppose I want to model airflow in a building. I can conjure up this building, and have a million data points and do that, but if you give me a bigger supercomputer, I’ll say: “You know what, I need to put in 10 million data points, I need to have higher resolution model of the building. You know, I need to model every doorknob and handle…” – there is no end to it! So at what point do you recognize that you are hitting diminishing returns? That you definitely need to model all the windows, but you don’t need to model every door handle or switch on the wall? Surely, everything has an impact on the overall airflow in the building, but those will probably be very minor. This is going to require expertise to determine at which point you stop.
So I can turn away and say, “Ah, they’re scientists and engineers, they will always ask for everything.”
But what happens if I ask a non-expert, a non-scientist? Sometimes they don’t know what digital transformation can do until they see it.
Shiva Dustdar How super is super enough? This is basically the question, right?
Rupak Biswas: From a pure bragging rights perspective, the Top 500 list is what we hear all about. And if you have a petaflop system, that will get you a lot of visibility in the community. Everybody will say: “Oh my God, the EU has this big supercomputer!”
But that may or may not translate into having a big impact. So one of the things we have always tried to do at NASA is tying it to an actual application impact. For example, using a larger supercomputer could provide more insight and ability to reduce airframe and engine noise that are particularly severe during an aircraft’s approach to landing. This is, in fact, a major focus of the design, retrofitting, and operation of current and future aircraft. Whatever your application is, of course, this goes back to the idea of translating abstract supercomputing power into real-world benefit that people can understand and appreciate.
Shiva Dustdar: Rupak, thank you for sharing those great insights and helping us understand how NASA makes the most out of its supercomputing investments. It is clear that much more education is needed across the board and especially in the financial community about the benefits of supercomputing to companies of all sizes. We hope that our study will further raise awareness and stimulate the development of new risk sharing products for the wider supercomputing ecosystem in Europe.