PRoductivity and Energy-efficiency through Abstraction-based Parallel Programming (PREAPP)

In today's exponential world of digital data, big data services have made the power consumption the lion's share of the total cost. For instance, Google data centers consume almost 260 MW, about a quarter of the output of a nuclear power plant, enough to power 200 000 homes. Energy efficiency is therefore considered a major criterion for "sustainable" computing systems and services over the data deluge. However, energy-efficient computing systems make parallel programming even more complex and thereby less robust due to requirements of massive parallelism, heterogeneity and data locality. The PREAPP project aims to devise novel programming models that will form foundations for a paradigm shift from energy "blind" to energy "aware" software development. The new models will enable one order of magnitude improvement in energy efficiency in comparison with today's multicore computing, thereby greatly advancing green computing and sustainable services. The new models will facilitate unprecedented productivity for implementing scientific big data applications that run effectively on large-scale high-performance computing (HPC) platforms, which are based on cutting-edge manycore architectures. The threshold of adopting large-scale parallel computing will thus be considerably lowered for a large number of computational scientists in several disciplines. The project has achieved its objectives. We have devised new models for designing energy-efficient parallel algorithms, new energy-efficient programming abstractions and frameworks, and new tuning approaches to optimize energy consumption for heterogeneous computing systems at runtime. To tackle the efficiency challenge for communication, we have devised new resource management approaches and scheduling algorithms to improve the efficiency of communication standards. We have developed high-level programming abstractions and frameworks that enhance productivity. We have exploited the new models and knowledge resulted from this project in several disciplines, including cyber-physical systems, cyber-security, and smart power systems. The PREAPP project team has established collaboration with the Computer Languages and Systems Software (CLaSS) group at Lawrence Berkeley National Lab. Together with CLaSS group, we have organized two international workshops at Simula and UiT and exchange visits with support from FRIPRO Overseas Research Grants.

Outcomes: The total number of scientific publications that related to this project is 22 of which there are 7 journal publications and 15 refereed proceedings papers. Besides, there are three technical reports of which one is being prepared for journal submission. Impacts: PREAPP research results have contributed to enhancing productivity, improving energy efficiency, and forming the foundations for a new energy-efficient computing paradigm. The new energy-efficient computing paradigm will enable highly efficient execution of code with minimal consumption of energy. With the current growth in large data centers and numbers of devices, this will have a considerable, long-term impact on environmental sustainability throughout ICT.

In today's exponential world of digital data, big data services have made the power consumption the lion's share of the total cost. For instance, Google data centres consume almost 260 MW, about a quarter of the output of a nuclear power plant, enough to power 200 000 homes. Energy efficiency is therefore considered a major criterion for "sustainable" computing systems and services over the data deluge. However, energy-efficient computing systems make parallel programming even more complex and thereby les s robust due to requirements of massive parallelism, heterogeneity and data locality. The PREAPP project aims to devise novel programming models that will form foundations for a paradigm shift from energy "blind" to energy "aware" software development. T he new models will enable one order of magnitude improvement in energy efficiency in comparison with today's multicore computing, thereby greatly advancing green computing and sustainable services. The new models will facilitate unprecedented productivity for implementing scientific big data applications that run effectively on large-scale high-performance computing (HPC) platforms, which are based on cutting-edge manycore architectures. The threshold of adopting large-scale parallel computing will thus b e considerably lowered for a large number of computational scientists in several disciplines. The PREAPP consortium consisting of world class universities (LIU, UiT) and research centres (INESC-ID, Simula), unites national and international experts with the required expertise to accomplish the ambitious but realistic goals of PREAPP.