EPCC Guest Lectures

EPCC Guest Lectures

Introduction

EPCC and the MSc in High Performance Computing run an annual Guest Lecture series, with speakers drawn from industry and academia. Talks focus on state-of-the-art applications of HPC in real world situations, with many reflecting the type of work our MSc students will undertake after graduation.

www.epcc.ed.ac.uk/msc

https://twitter.com/#!/EPCCedinburgh

How to Use

This page shows a list of up-to-date episodes for this channel. You can play individual episode by pressing the 'podcast' button next to each episode, or you can subscribe to this channel and be kept up-to-date with new episodes.

Drag this icon to your aggregator to subscribe Click on this link to launch iTunes and subscribe to this channel (requires iTunes)

If you have any feedback (or problems) regarding this channel then please contact Iain Bethune.

Podcast Episodes

Play (left-click) or download (right-click then 'Save As...')
[Fri, 21 Mar, 2014]
Adventures in the Biology trade : Bioinformatics in the Petabyte era (60 mins, ~42 MB)
Description:
Bioinformatics and more widely Computational Biology is a largely data-driven Science. The array of high-throughput technology platforms in the last 10 years mean that the amount of data being generated in this field is likely to enter into Exabytes by 2020. The challenges associated with this are quite different from the data sets generated by High Energy Physics or Astrophysics in that they tend to gathered from a wide variety of different providers. Meta-analyses of these data sets can give startling new insights but come with many caveats - in particular that the quality of the data from each provider can be highly variable. I will spend some time talking about one set of experiences I have dealing with one specific technology platform and in particular how it is clear that the detection of bias in data sets is a key element of any high-throughput analysis. This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 28 Feb, 2014]
High Performance Computing - Driving innovation and capability: Ian Wardrope, Intel (55 mins, ~4 MB)
Description:
Intel will provide an insight into future HPC technology development looking at hardware trends, ecosystem support and the challenges around ExaScale computing. The talk will also touch upon the convergence of High Performance Computing and High Performance Data Analytics, examining where the effective use of this rapidly maturing capability can provide industry and academia with a competitive advantage. This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Primegrid_logo
Play (left-click) or download (right-click then 'Save As...')
[Sat, 07 Sep, 2013]
PrimeGrid: Join in the search for a world record prime number (45 mins, ~41 MB)
Description:
PrimeGrid is a volunteer computing project that gives participants the chance to be the discoverer of a new world record prime number! In addition, we are working towards the solution of several mathematical problems which have remained unsolved for over 50 years. The talk will cover some basic facts about prime numbers, the history of the search for large primes (and a little of the maths!), and show the audience how they can use their computers to join PrimeGrid and find new primes of their own. This talk was given at the British Science Festival on Sat 7th September 2013 by Iain Bethune of EPCC.
Play (left-click) or download (right-click then 'Save As...')
[Tue, 02 Apr, 2013]
The Power of Many: Running Many Simulations on Many : Dr. Shantenu Jha, Rutgers University (57 mins, ~28 MB)
Description:
There are several important science and engineering problems that require the coordinated execution of multiple high-performance simulations. Some common scenarios include but are not limited to, "an ensemble of tasks", "loosely-coupled simulations of tightly-coupled simulations" or "multi-component multi-physics simulations". However, historically supercomputing centers, have supported and priortised the execution of single "jobs" on supercomputers. Not suprisingly, the tools and capabilities to support coordinated multiple simulations are limited.
A promising way to overcome this common limitation is the use of a Pilot-Job --- which can be defined as a container or placeholder job to provide multi-level scheduling via an application-level scheduling overlay over the system scheduler. We discuss both the theory and practise of Pilot-Jobs: Specifically, we introduce the P* Model of Pilot-Jobs and present "BigJob" as a SAGA-based extensible, interopable and scalable implementation of the P* Model. We then discuss several science problems that have/are using BigJob to execute multiple simulations at unprecedented scales on a range of supercomputers and distributed supercomputing infrastructure such as XSEDE. This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 22 Mar, 2013]
Games: Playing with Threads: Ben Nicholson, Frontier Developments (53 mins, ~21 MB)
Description:
In this talk I will give a brief history of parallel processing in games and how the industry has responded to hardware changes in its constant race to create games with more, better and faster. I then consider some of the lessons we have learned so far and finish with my opinion on how a future game engine might be structured to target many-core architectures. This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 15 Mar, 2013]
Fast as Lightning: High Performance Computing at the Met Office: Dr. Chris Maynard, Met Office (59 mins, ~27 MB)
Description:
The Met Office hosts some the largest computers in the UK to predict the weather and changes in the climate. Best known for the Public Weather Service, the Met Office also advises UK government and other organisations world wide. The difficulties of modelling such a complex system as the Earth's atmosphere, how it is actually done and what future computational challenges lie ahead will be discussed. This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 01 Mar, 2013]
Data Centric Deep Computing: Dr. Crispin Keable, IBM (59 mins, ~27 MB)
Description:
This talk will look at the drivers leading the way to Exascale computing. It will examine technology trends, and use these to infer some of the characteristics of future large scale HPC systems. It will also look in to what this means for the software environment, reliability, use and affordability of IT systems at the high end. Recently, the IT industry has become more focused on problems involving the use and management of large data sets (Big Data) and we map this trend on to the consequences of Exascale computing. What will future architectures look like, how will they be used, and what does that mean for applications development today? This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 23 Mar, 2012]
FireGrid - A better response to fire emergencies: Dr. George Beckett, EPCC (42 mins, ~2 MB)
Description:
Building design has been revolutionised in recent years. New materials and better construction techniques have allowed bespoke and impressive public spaces to be created. Consider the Millennium Dome, Hong Kong airport, or the 2012 Olympic stadium.

These bespoke spaces represent a significant challenge for fire safety. Unlike for conventional buildings, with regular layouts and dimensions, there is very little experience to suggest how fires might develop in these spaces, and how people might react in an emergency situation.

An Edinburgh-led consortium, called FireGrid, is developing technology designed to tackle this specific problem head-on. By combining advanced modelling with sensor data, in a high performance computing (HPC) context, the FireGrid team are developing a system that can provide fire fighters with better information to support their response, to control the impact of a fire emergency, and to reduce the risk to life and property.

In this talk we will review progress of the FireGrid consortium and describe a technology prototype that has been built to demonstrate the potential for such a system, and to stress some of the key technologies that are required.
This talk was given as part of our MSc in HPC's 'HPC Ecosystem' course.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 02 Mar, 2012]
GPU Computing: Past, Present and Future (47 mins, ~21 MB)
Description:
The past five years have seen the use of graphical processing units for computation grow from being the interest of handful of early adopters to a mainstream technology used in the world’s largest supercomputers. The CUDA GPU programming ecosystem today provides all that a developer needs to accelerate scientific applications with GPUs. The architecture of a GPU has much to offer to the future of large-scale computing where energy-efficiency is paramount. NVIDIA is the lead contractor for the DARPA-funded Echelon project investigating efficient parallel computer architectures for the exascale era.

Timothy Lanfear is a Solution Architect in NVIDIA’s Professional Solutions Group, promoting the use of the NVIDIA Tesla(TM) computing solution for high-performance computing. He has twenty years’ experience in HPC, starting as a computational scientist in British Aerospace’s corporate research centre, and then moving to technical pre-sales roles with Hitachi, ClearSpeed, and most recently NVIDIA. He has a degree in Electrical Engineering and a PhD for research in the field of graph theory, both from Imperial College London.
Play (left-click) or download (right-click then 'Save As...')
[Fri, 14 Oct, 2011]
Prime numbers, large integer arithmetic and a 2PF task farm: Iain Bethune, EPCC (43 mins, ~21 MB)
Description:
We will review some basic properties and theorems regarding prime numbers, and take a quick trip through the history of prime number searching. Secondly, we will discuss two classes of algorithms of importance for computational primality testing - Sieving and the Lucas-Lehmer (and similar) tests - and their implementations on modern CPUs and GPUs. Finally, we will introduce GIMPS and PrimeGrid, two large and well-known distributed prime search projects.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Wed, 16 Mar, 2011]
An introduction to distributed version control systems: Lawrence Mitchell, EPCC (60 mins, ~54 MB)
Description:
There is a plethora of version control systems and it is not obvious which to choose. Do you pine for the warm, comforting blanket of CVS? Fear not! I shall give an overview of the zoo of version control systems available to the modern programmer. I'll describe the design principles (there is much overlap) and how these influence the use patterns. I hope to convince you that modern systems are at least as good as CVS and Subversion, and to indicate some of the areas in which they make a programmer's life much, much easier.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 11 Feb, 2011]
Building Cloud Infrastructures (looking inside Flexiscale) (63 mins, ~29 MB)
Description:
Cloud computing has become one of the most advertised and talked about technologies for the past few years since many ISPs/IT firms have begun to adopt virtualisation technologies. As cloud technologies mature, more and more businesses are moving their services into “the cloud” or building internal clouds to cater for their IT needs. Gihan Munasinghe & Tabassum Sharif of Flexiant discuss the advantages of having cloud infrastructures, introduce Flexiscale and look into how you can build your own internal cloud infrastructure.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 28 Jan, 2011]
Fluidity: an adaptive, unstructured mesh CFD framework. Applications, software development methods and parallel performance: Dr Jon Hill, Imperial College London (48 mins, ~22 MB)
Description:
Fluidity is a powerful Computational Fluid Dynamics framework developed at Imperial College over the last 20 years. Fluidity can be applied to a number of scientific applications, ranging from classic CFD, to oceans and multi-material problems. It uses a variety of discretisations on an unstructured mesh and includes a number of novel and innovative features, such as adaptive re-meshing, a user friendly interface, and a python interface. The main advantage Fluidity has is the adaptive re-meshing, which allows resolution to be focused where and when it is required during a simulation, depending on the simulation state at that time. This talk will briefly cover some of the features and science behind Fluidity before covering details of AMCG’s software development methodologies and the parallel scaling behaviour of Fluidity on HECToR. Finally, future developments will be covered.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 21 Jan, 2011]
Applications of HPC to Materials Chemistry: Dr. Scott Woodley, University College London (55 mins, ~38 MB)
Description:
The Materials Chemistry Consortium (MCC) is the single largest consumer of resources on the HECToR National Supercomputer. In this talk I will give an overview of how MCC members exploit HECToR in a wide range of materials chemistry research and the types of applications that we are interested in.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 19 Mar, 2010]
High Performance Storage with SSDs: Barry Whyte, IBM UK (51 mins, ~47 MB)
Description:
Traditional disk technology has lagged behind processor and memory technology for some time. During this session we will discuss what the problems are and how new technologies such as enterprise class Solid State Drives (SSD) are helping. Until now, high performance storage systems have been designed around the characteristics of spinning magnetic media. This session will also discuss design considerations for such high performance systems, an overview of storage virtualization, and how these concepts are increasingly important with the move towards solid state storage media.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 26 Feb, 2010]
Lightning Fast! HPC at the Met Office (63 mins, ~57 MB)
Description:
The Met Office has long used high performance computing to support its activities in weather prediction and climate research. This talk starts by giving an overview of the numerical weather prediction process and continues with a look at the HPC platforms used to make solving these problems possible. Some of the challenges currently faced, particularly in the area of scalability, are presented and recent progress in these areas discussed.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 05 Feb, 2010]
Instruction Level Parallelism - a historical perspective: Prof. Roland Ibbett, EPCC (70 mins, ~64 MB)
Description:
This talk traces the development of the different types of instruction-level parallelism that have been incorporated into the hardware of processors from the early 1960s to the present day. We will see how the use of parallel function units in the CDC 6600 eventually led to the design of the Cray-1 with its vector processing instructions and superscalar processors such as the Alpha. The talk also covers Very Long Instruction Word systems and SIMD systems. The talk is accompanied by visual demonstrations of the activities that occur within processors using simulation models developed using the HASE computer architecture simulation environment.
Links:
Play (left-click) or download (right-click then 'Save As...')
[Fri, 22 Jan, 2010]
Lattice QCD, HPC and GPUs: Dr Alistair Hart, University of Edinburgh (63 mins, ~58 MB)
Description:
Lattice QCD is a very important tool for understanding the Standard Model of particle physics. It presents enormous computational challenges, both in carrying out simulations and in making sure they are as efficient as possible. This lecture explains why we need lattice QCD; describes how the UKQCD Collaboration sets out to overcome these computational challenges; and illustrates how GPUs are used here in Edinburgh.
Links: