HPC: from supercomputers to supercomputing
4 May 2018 | 0
The area of high-performance computing (HPC) used to be all about the machines – vast beasts with thousands of cores drawing massive power. Cray, IBM, Fujitsu and SGI were common names in the Top500 list of the most powerful supercomputers, with China now a regular entrant.
However, according to Prof JC Desplat, director, Irish Centre for High-End Computing (ICHEC), supercomputing is maturing beyond the hardware.
Politicians are beginning to see HPC in terms of driving competitiveness, says Desplat.
“HPC is becoming more about the knowledge of computing for performance, not just supercomputing,” Prof JC Desplat, ICHEC
Whereas previously it was only seen as an academic research tool, he says, it is now properly considered beyond this for the likes of digital transformation in government services.
He also points out that it has become a key European focus as part of the Digital Agenda. EuroHPC, the European Union HPC initiative, has seen its seven initial signatories joined by a further eight, as of February of this year.
European Digital Day
The recent European Digital Day in April also brought together high-level stakeholders in the fields of digital technology and telecommunication, and was organised by the Bulgarian Presidency of the Council of the European Union.
The event followed-up on agreements that had been reached on the previous Digital Day 2017, that concluded agreements and actions related to HPC, digital transformation of jobs and skills, digitisation of industry, and connected automated mobility.
The HPC focus, said Desplat, is around creating high value jobs in areas of study such as climate change research, medical diagnostics, social and urban planning, environmental studies, etc.
Recent changes in HPC, however, have allowed the focus to broaden beyond the machines and into the area of efficiency and improving utilisation.
More energy efficient machines, as well as more accessible usage, has allowed many more areas of research to be opened up to HPC.
The United States has stated ambitions to develop an Exaflop supercomputer, but Desplat points out that with current technologies, such a machine might cost €100 million a year in electricity costs alone.
This has prompted development in other directions, such as more energy efficient cores, such as ARM derived chips, but also refinement and re-engineering of software and applications to better utilise existing, and make more efficient future machines.
HPC is becoming more about “the knowledge of computing for performance, not just supercomputing,” says Desplat.
The ICHEC is working with academia, industry and government for the democratisation of supercomputing that has arisen from these changes.
The centre won a tender for the redevelopment of the Met.ie site, and the forecasting capabilities behind it.
This required not just a new system for forecasting capable of a 9km resolution with a 10-day outlook, but also high availability and the capability to handle surge demand, such as during the recent red-level weather events.
The project, said Desplat, characterises the maturing HPC scene, where the horsepower of supercomputers is giving way to large, complex data applications where methodology takes precedence.
Ireland is well placed to take advantage of this development, he assures, but there are areas to be addressed. The country currently has no national data management infrastructure, he warns, which is important as to properly develop machine learning (ML) requires large volumes of data which need to be properly managed, controlled and distributed.
However, Ireland is already a key contributor to the education and training side of HPC, as key staff members of ICHEC, such as Simon Wong PhD, Education, Training and Outreach, Lead Computational Scientist, ICHEC, work both locally and with international bodies to develop programmes for scientists, researchers and developers.
Wong describes how work with the institutions such as Beaumont Hospital and the Royal College of Surgeons Ireland (RCSI) has seen projects such as the application of ML algorithms to the brain scans of epilepsy sufferers. This has allowed the work of clinicians to be supplemented, rather than replaced, a kind of force multiplier to provide support and second views.
A lot of the work is about optimising and refining for performance, said Wong, and learning how to use resources more efficiently.
Wong is heavily involved in developing and running HPC boot camps, with organisations such as the Partnership for Advanced Computing (PRACE). These efforts are about training and programmes to allow people to be prepared to use machines and deploy code.
This democratisation process has allowed supercomputing to be applied in ways and areas not necessarily viable before, such as agriculture, environment monitoring, urban planning and more.
Wong said that they are working to combine the experience of teams all over Europe, to cover requirements, and allow them to work more efficiently with the resources.
Alastair McKinstry, Environmental Activities lead, ICHEC, said that changes in the technologies, such as cloud computing, have opened up new ways of processing data, adding further value to the work of data gatherers, such as satellite data from European Space Agency (ESA) programmes Sentinel and Copernicus.
HPC processing of satellite data, from land survey and ice sheets to weather patterns and agricultural activities and deforestation, said McKinstrey, have real world impact.
But also, the results of such processing, can be opened up to organisations for commercial use. Image processing has been changed by HPC, says McKinstrey. Old, one at a time, human-centric processes have now become algorithmic, with analysis from the cloud. ML applied can be applied to mass data sets and is an opportunity to tune and refine their operation.
Cloud availability for both storage and processing power mean that any organisation can engage with data without huge outlay in terms of storage or processing power. This will bring a new generation of data-based services in much the same manner as GPS technology spawned.
With the trend away from the machines and more towards the discipline of supercomputing, Desplat is unequivocal in his view.
“HPC is an engine of Digital Transformation,” he stated.