The UrgentHPC workshop, taking place Sunday (Nov. 17) at SC19, is focused on using HPC and real-time data for urgent decision making in response to disasters such as wildfires, flooding, health emergencies, and accidents. We chat with organizer Nick Brown, research fellow at EPCC, University of Edinburgh, to learn more.
HPCwire: Tell us about the mission and background of UrgentHPC?
Nick Brown: We started from the observation that our ability to capture data is growing at an incredible rate, as is the power of HPC machines. But at the same time, it seems that there is always some sort of disaster in the news, from wildfires, to health emergencies, to extreme weather, and it seems to me that the frequency and severity of these things is only getting worse! So I think it is natural to ask ourselves as a community, whether there is a role that we can play in helping to tackle these emergencies. HPC has been able simulate disasters after the fact for many years, but what about if we could run these urgent simulations in real-time, fed by the latest data streaming in from the field? What sort of opportunities could this then open-up to assist urgent responders, ultimately translating into more lives saved and reduced economic impact?
What I find so fascinating is not only the significant potential impact if we can get this right, but also all the technical challenges that must be overcome to reach this point! These challenges cover many aspects of our and other communities. From visualisation, to data engineering, HPC system support, algorithmic techniques, and integration with current disaster response systems. One such example are the batch queues employed by all major HPC machines, as these are simply not set-up to support an urgent workload – it’s pointless waiting two hours for a job to run when the forest is burning right now!
HPCwire: What is urgent decision making?
This is where a front-line responder is making real world decisions in response to some emergency, for instance the recent California wildfires. With lives often in the balance, it is crucial for these individuals to make correct decisions first time, every time and a key question for us, is how real-time data and HPC can assist here.
HPCwire: What are some of the use cases UrgentHPC is focused on?
From discussions with the community it surprised us how much is going on and how many groups would like to use HPC, to some extent or another, with their disaster response applications. For instance, the keynote speaker at our workshop this year is from Technosylva, a company who develop the world’s leading wildfire simulation code. This has been used extensively during the recent Californian wildfires and we are really excited that Joaquin will be joining us to talk about their work and the critical role played by supercomputing, along with future plans to take advantage of the even greater capability that is being developed by the HPC community.
Other individuals involved in the workshop are interested in areas including mosquito borne diseases, space weather anomalies, earthquakes, tsunamis, wildland fire and smoke progression. So really wide range of use-cases, and we are hoping to identify even more during the workshop session!
HPCwire: There’s a workshop this year at SC — what is on the agenda?
Yup, and it is great to be back at SC as this initiative began a year ago with a BoF at SC18. At the time I was pleasantly surprised by how many people were interested in this topic and seemed to be working on related activities. Following on from this we felt that it would be beneficial to pursue developing a community around urgent computing and HPC, and potentially consolidate efforts, hence this workshop!
We have a fantastic programme (https://www.urgenthpc.com) and I am really excited for the workshop which is running on Sunday afternoon (17th) from 2:30pm in room 603. I mentioned previously about the wildfire keynote talk, we also have six research papers being presented which all describe solutions and technologies for addressing different parts of the overall urgent HPC challenge. In addition we also have a panel session with some really interesting individuals lined up who will provide their own views on the use of HPC for urgent decision making and respond to some of the themes raised in the workshop.
HPCwire: What kind of computing infrastructure is required and how would that be delivered (on-prem, in the cloud)?
That’s a very interesting question, not least because it was one of the most extensively discussed topics during our BoF last year! It is clear to us that HPC machines as they currently stand are not enough (not just issues around the suitability of the batch system, but furthermore supercomputers often do not operate to the SLAs required by emergency responders). Certainly the cloud has a role to play, but it is not a silver bullet because, whilst there is some elasticity here, being able to unpredictably run jobs immediately requiring high performance across thousands of nodes is also not really a usage model these organisations have considered.
Inevitably this is going to require further development, both on the technology and policy sides, across a wide variety of infrastructure and we also believe that there is a role for edge computing to play, where computation (such as data reduction) can be quickly performed at source to reduce the amount of work required centrally.
Nowadays there are quite a lot of sensors, for instance satellites, out there that freely available data, which we think helps on the infrastructure side quite a lot too, although of course the pipes must be in place to transmit it quickly enough for processing.
HPCwire: Seems like there is a big data science element here. Is there focus on a converged stack? Bringing HPC and big data tools together?
You are absolutely right, and I think some of the opportunities that are opening up are down to some of the efforts made around this convergence already. Although there is still quite a way to go – an example of this is the Topology ToolKit (TTK) developed by Sorbonne University in Paris. This is an advanced method of feature extraction which can be run on the raw data output of HPC codes, and significantly reduces the amount of data that must then be stored or transferred, but the big challenge is that currently it does not support running across multiple nodes. So we think things like this, combining the knowledge and algorithms of other communities with our expertise in parallelism, could be a very potent collaboration.
HPCwire: What (other) challenges exist on the technology and policy side?
Lots, and I suspect even more will be uncovered at the workshop! As a technologist myself, I tend to focus on that side, but do think the policy side is at-least as challenging or more. This is because fully realising the use of HPC for urgent decision making will require changes to how HPC machines are operated and used. This is one of the reasons why I think activities like our workshop are so important, as if we can bring together the community and build momentum, then it will make a much stronger case to the machine owners and operators to change things. Bearing in mind some of these rules and policies have been around for decades, there is an uphill struggle!
HPCwire: What is your role and how did you come to be involved?
I first got involved in this via the VESTEC (https://vestec-project.eu/) EU funded FET project, where we are looking to fuse HPC with real-time data for disaster response. This project involves partners from across Europe with a wide variety of expertise, from fire simulation, to in-situ visualisation, to HPC. But what’s important to say is that we appreciate that to build a community one needs to look outwards and consider the global picture, especially as there is some interesting work going across the US and further afield. This is represented in my fellow organisers of the workshop who are from PNNL, NCAR, and the LEXIS EU project, and we think this mix is a strong one in encouraging good participation across the board.
HPCwire: Who is the workshop for? Who should attend?
Anyone with any interest in the topic will be more than welcome! It’s funny, the more we chat with people about this, the more we discover just how much is going on that is complementary to the central aims of what we are trying to do. So irrespective of your area of expertise, please come along and let’s see what we can learn from each other! As a frequent attendee of SC I am really excited both to have the opportunity to run a workshop at the conference and also for the conference itself – it’s going to be a great week ahead!
About Nick Brown
Dr Nick Brown is a research fellow at EPCC, University of Edinburgh, with interests in data engineering, machine learning, parallel programming language design, novel compute architectures, compilers and runtimes. He is a work package leader in VESTEC, an EU FET HPC project, which aims to fuse HPC with real-time data for urgent decision making and as part of this is responsible for the HPC side of this project. He has worked with both industry and academia, for instance managing a project using machine learning to optimise the interpretation of well log data in the oil and gas industry. He has worked on a number of large scale parallel codes including developing MONC, an atmospheric model used by the UK climate and weather communities which involves novel in-situ data analytics. Nick is a course organiser on EPCC’s MSc in HPC and data science courses, and supervises MSc and PhD students.