#	Pagina
attuale pagina	/open-h2020/projects/213151/results.html

Report

Teaser, summary, work performed and final results

Periodic Reporting for period 1 - LEGaTO (Low Energy Toolset for Heterogeneous Computing)

Teaser

Summary

Imagine being able to write an application once and have it scheduled from Internet of Things (IoT) sensors to data centres and supercomputers, all while making the most energy-efficient use of the hardware available. This is the main challenge the European project LEGaTO - Low Energy Toolset for Heterogeneous Computing - is tackling over three years of research and development.
Why the emphasis on energy? Information and Communications Technology (ICT) has made great strides in recent decades, impacting the quality of life of every European citizen. However, this transformation comes with a cost, currently ICT sector accounts for 10% of the European electricity consumption. Furthermore, this percentage is expected to increase as the industry is hitting fundamental physics limits to decrease the size of computational building blocks, thus making them less energy efficient. This situation clearly requires a fundamental energy-savings approach at all layers of the computational stack. While attaining these energy-savings, care must be taken not to compromise the system from security attacks, and the system should continue to function correctly in spite of occasional faults. Finally, the energy-efficient software stack must be easy to program and to maintain.
The overall objective of the LEGaTO project is to produce a mature software stack to optimize the energy-efficiency of heterogeneous computing. The project will strive to achieve this objective through employing a task-based programming model, coupled to a dataflow runtime while simultaneously ensuring security, resilience and programmability. The LEGaTO project will apply this energy-efficient software stack for heterogeneous hardware to the use cases of health care, smart home/city and machine learning.
The LEGaTO project is funded by the European Commission with a budget of more than 5 million euros and will run from December 2017 to November 2020. The partners of the project are Barcelona Supercomputing Center (BSC, Spain), Universitaet Bielefeld (UNIBI, Germany), Universite de Neuchatel (UNINE, Switzerland), Chalmers Tekniska Hoegskola AB (CHALMERS, Sweden), Machine Intelligence Sweden AB (MIS, Sweden), Technische UniversitÃ¤t Dresden (TUD, Germany), Christmann Informationstechnik + Medien GmbH & Co. KG (CHR, Germany), Helmholtz-Zentrum fÃ¼r Infektionsforschung GmbH (HZI, Germany), TECHNION - Israel Institute of Technology (TECHNION, Israel), and Maxeler Technologies Limited (MAXELER, United Kingdom).

Work performed

Results:
In an early success story for LEGaTO, partner MIS has spun-off a new company, EmbeDL AB, to commercialize the LEGaTO developed Deep Learning energy optimiser tool with the single focus on taking this technology to the market. The technology will be sold on a license based revenue model. EmbeDL AB technology is already under consideration with customer pilots in Volvo, Zenuity and Veoneer. Also EmbeDL AB has been in discussions with other sectors, e.g. Telecom and Life Sciences.
The project has developed several software packages and tools in the first phase of the project and ten of these packages have been made available for the public through the project Software Components portal.
Furthermore, partners Maxeler and HZI were able to increase the energy-efficiency of a healthcare pilot application kernel by two orders of magnitude. On the research side, partner BSC has decreased FPGA memory energy consumption by one order of magnitude through a technique termed undervolting; the results of this work appeared in the 51st Annual Symposium on Microarchitecture (MICRO) which is one of the top computer architecture conferences where typically only one in five submissions are accepted.

Final results

Progress beyond SoA and Expected Results:
The slowdown of the Mooreâ€™s Law has increased the importance of specialized hardware accelerators such as Graphical Processing Units (GPUs) and Field Programmable Gate Arrays (FPGAs) in mainstream computing services. These accelerators supplant the traditional way of squeezing more performance out of CPUs (i.e. shrinking the transistor size through the Moore Law). Moreover, different types of accelerators are used together in heterogeneous computing systems. Companies such as Intel has started to build CPUs and GPUs on the same chip and companies such as Amazon has started to offer the renting of FPGAs in their cloud computing platforms. While these systems are potentially more energy-efficient, they require a tight integration between hardware and software as well as complete support of the software stack, otherwise they do not fulfill their potential and will be wasteful of energy. This support has been mostly lacking so far. The primary ambition of the LEGaTO project is to start with a Made-in-Europe mature software stack, and optimizing this stack to support energy-efficient computing on a commercial cutting-edge European-developed CPUâ€“GPUâ€“FPGA heterogeneous hardware substrate, which will lead to an order of magnitude increase in energy efficiency.
Impact:
The LEGaTO project will demonstrate the productivity for developing applications for two use cases from outside the traditional high-tech sector: Healthcare and Smart Home/City. The LEGaTO low-energy optimizations are mostly â€œunder-the-hoodâ€ at the runtime level, therefore the application programmer does not need to become a low-energy expert to make her application low-energy. One of the main goals of LEGaTO is to support a â€œwrite once run everywhereâ€ programming paradigm through employing a task-based dataflow software ecosystem. In turn, this paradigm is instantiated by seamlessly moving tasks to the most energy-efficient hardware substrate; without requiring any effort from the programmer/developer. The project has already demonstrated substantial decrease in energy consumption in a pilot healthcare application through the use of the LEGaTO toolset. LEGaTOâ€™s target is to achieve similar energy savings for the Smart Home use-case so that assisted living scenarios such as the LEGaTO Smart Mirror application (see Figure) becomes possible to deploy in senior citizen homes.
The LEGaTO consortium is highly committed with the involvement of SMEs, in fact, 3 of the 5 non-academic partners in the consortium are SMEs or mid-caps. The role that SMEs play in the LEGaTO Project is critical: MIS will not only develop the Machine Learning use-case, they will also provide know-how to other partners who will apply Machine Learning schemes. CHR will be working on increasing the value of LEGaTO hardware through the integration of a high-performance interconnect fabric. Maxeler, the third SME, will be providing the other LEGaTO initial platform. With respect to facilitating SMEs competitiveness, the consortium is fully aware of the barriers that SMEs occasionally encounter in funding their technology platforms. To correct this direction, the LEGaTO ecosystem will enable the use of cost-effective low-energy distributed computing solutions that provide a substantial percent reduction in the Total Cost of Ownership (TCO), i.e. the cost to buy, own, operate, and manage; when compared to the systems currently on the market. The SMEs will highly benefit from the reduction of such costs. In addition, open source software provides a low cost entry point for SMEs and startups, allowing them to lower the risk and increase the speed of European business innovations.