University of Glasgow projects set to tackle many core computing challenges

The projects have received more than £3m in funding from the Engineering and Physical Research Council’s (EPSRC) System Approaches to Distributed and Embedded Architectures (SADEA) project. Across the UK, a total of five SADEA projects will develop new software to help unlock the full potential of manycore computers.

Until recently, advances in computing power relied on increases in processor clock speeds, which allowed more computations per second. However, the physical limitations of silicon manufacturing led to a gradual plateauing of clock speeds. To get around the problem, processor manufacturers began to place multiple processing cores on the same silicon wafer. These ‘multicore’ systems are now commonplace in devices including laptop computers, tablet PCs and mobile phones. ‘Manycore’ systems, which incorporate hundreds of processors or more, are now becoming increasingly common for large-scale computing projects.

However, the development of software properly tailored to the challenges of multi and manycore systems has lagged behind hardware development, meaning that many of the potential advantages of new hardware architecture have not yet been fully exploited.

One project, led by Dr Wim Vanderbauwhede, will focus on creating a software ‘Rosetta Stone’ to allow existing computer programs to benefit from the power of manycore system processing without requiring programmers to rewrite their code from scratch. Over the next five years, Dr Vanderbauwhede and his collaborators at Imperial College London and Heriot-Watt University will develop a compiler system to allow programs to take full advantage of manycore systems as well as hybrid systems which also incorporate graphics processors and field-programmable gate arrays (FPGAs). The project could be of tremendous benefit to complex computational tasks such as weather forecasting, which rely on custom software which may not be easily optimised to run on manycore systems.

A second five-year project, led by Dr Jeremy Singer, is aiming to develop ‘AnyScale’ technology which will allow applications to intelligently distribute their computational requirements across a wide range of local and wirelessly-connected hardware. The first project will support the development of a movement controller app to allow a bipedal robot to traverse a wide range of terrain. Walking upright on two legs is a complex computational task which is currently extremely challenging to achieve using only the limited processors built in to robots. The app, developed in collaboration with the Universities of Edinburgh and Manchester, will allow the necessary computing to be done both on-board and wirelessly via servers or cloud-based systems and could lead to improved robot explorers capable of better exploring disaster zones or other environments which are challenging for traditional human exploration.

The third project, headed by Professor Phil Trinder, will last for three years and will develop a reliable method to ensure the performance of software is consistent across a wide range of multicore systems, from desktops to high-end supercomputers. Software performance is often compromised by a program’s inaccurate assumptions about the architecture of the hardware it is running on. Professor Trinder is working to develop a new form of compiler, using dynamic scheduling and just-in-time compilation techniques, to allow software to run at its maximum effectiveness without forcing programmers to consider the different demands of every possible current and future hardware configuration.

Professor Joe Sventek, Head of the School of Computing Science at the University, said: “We’re proud to be the leading partner in three of the five SADEA projects funded by the EPSRC. It’s a tremendous endorsement of our academics’ skillset and recognises the research expertise of the members of our Glasgow Parallelism Group. We’re excited about the work and looking forward to seeing the results of the projects over the coming years.”