In 2012, the ICT KTN begins a major series of webinars on Multicore, working together with major vendors, such as AMD and Intel, to show how we can get the maximum advantage from this evolving technology. ICT KTN’s Director of Cloud Computing and Government IT, Ian Osborne introduces the series:
The consequence of Moore’s Law, putting more transistors on a chip, has been a 2x increase of compute capacity every 18 months. This has come at a cost: increased power consumption, heat generated and air conditioning requirements are needed to compensate. About 5 years ago the heat being generated passed the point of practicality (too hot to be used without extreme cooling solutions) and dual core processors where introduced. In principle, multicore computers (now up to 8 processors on a general purpose computing device) have kept alive the doubling of compute power every 18 months. However, practically speaking, the software industry does not have the tools and flexibility to fully utilise these new systems. This has resulted in reduced utilisation, at reduced power – which is good, below the levels that we could achieve with the last generation of single core servers. For example, a modern quad core server may only be utilised at an average of 18% per core because of shared memory and I/O limitations. Our most popular tool for utilising multi-core configurations is virtualisation and the deployment of virtual machines on each core. However, this is most vulnerable to shared memory problems and performance is impacted.
What is needed is a fresh approach to planning and deploying applications to take advantage of a large amount of distributed processing, allowing for parallel execution, reducing time taken to deliver results. This can most effectively done in applications like spreadsheets where calculations can take place in parallel based on incoming data, and least usefully done where database entries have to be queried and updated in series, such as where people are buying items from stock, and that stock level must be accurate. With single core processing, applications could be written as a single line of logic for both cases, it would take longer, but there would be no benefit to breaking the tasks up. In parallel applications the opposite is true and we are heading into a new world where the demands of multiple users must be reflected in design for integrity as well as design for minimum speed and resource consumed.
There are now a large number of multicore devices in production, from all vendors, and we are anticipating the introduction of devices with the capacity of 16, 32, 64 and even a 100 cores in the future. How will our serial thought processes have to change to design systems which will operate in such environments? Will our existing tools and methods scale to meet these needs? What are the new ideas which can help release the power of such devices, and perhaps help the industry adapt to the requirement for low power, highly efficient computing systems? The ICT KTN is engaged in the process of shedding light on these challenges. In 2012, a webinar series “Maximising the Multicore Advantage”, will explore the architecture and design of the new generation compute devices and the tools available to maximise their use. The Technology Strategy Board has introduced a cohort of Knowledge Transfer Partnerships, aimed at transferring knowledge and best practice from academia to industry, starting with the volume end of the market with AMD on 11th January and Intel in February.
The complete list of vendors with whom we are planning to collaborate includes the "who's who" of chip vendors from the top CPU and GPU suppliers, which also include ARM, IBM, nVidia and Oracle/Sun. Each webinar will be free to access "live" on the day or archived via the ICT KTN website. We hope to feature a webinar towards the end of the series which will highlight the programming issues associated with larger numbers of cores describing strategies for their management.
In September we will be holding a major Multicore event with partners, Verification Systems Limited (TVS), in Bristol, where we will highlight the new approaches to programming. This is a follow up to to the highly successful Programming Multicore II event held last September and will feature many of the contributors to this series as well as other suppliers with relevant offerings for the market.
This comprehensive programme should appeal to programmers, designers, technologists and technology leaders with the goal of delivering new applications and services in computing environments from mobile platforms through to global distributed systems.
We would urge you to join the ICT KTN as it is a key vehicle for ongoing discussion in the community, within the main group there is also a Scalable Computing area. Further opportunity for debate also exists on the Multicore LinkedIn Group.
Through this series of activities we expect to shed light on the challenges that multicore programming presents and the opportunities for innovation through lower energy use and higher performance which are truly valuable in today's applications systems.
Find out more about the Maximising the Multicore Advantage series of ICT KTN webinars.