SHiPS wants to encourage the exchange of ideas and, possibly, build a community of researchers working on sustainable HPC algorithms, platforms, programming paradigms, and theoretical models.
Nowadays, sustainability is becoming a key challenge in science as in many human activities. As it is declined in the 2030 Agenda for Sustainable Development, adopted by United Nations member states in 2015, the term intertwines the need for a more socially just society with the protection of planetary boundaries. During the last decades, besides the natural strive to improve the system’s performance, the High-Performance Computing (HPC) research field has also deeply focused on enhancing the energy efficiency of supercomputers’ hardware. Since nowadays HPC machines have hundreds of thousands of CPUs, reducing energy consumption by just a few Watts per core can easily produce significant savings in terms of power and, ultimately, money. The huge interest in this area is testified above all by the always-increasing popularity in the Green 500 list of energy-efficient supercomputers.
Nonetheless, sustainability does not limit its semantics to the environmental impact of the development but also includes economical and social aspects.
In a large computing system, sustainability can be achieved at different levels. A mandatory step towards this ambitious goal is investigating how individual components (from low-level hardware constituents to nodes, infrastructures, operating systems, and high-level software) contribute to the whole system's footprint. For example, now that HPC hardware plays a key role in many fields--being extensively used in a plethora of applications ranging from scientific software to Artificial Intelligence (AI) algorithms--it is becoming more and more important to provide means to estimate the power consumed by applications. More precisely, it is becoming increasingly pressing the need to take into account also the relation between the characteristics of the parallel program being executed, and the power consumption and computational costs.
We believe that HPC systems shall play their part in the achievement of the ambitious goal of sustainable development, not only by investigating ways to reduce their environmental impact at the hardware and infrastructural level, but also by studying the sustainability of high-performance algorithms, proposing methods and tools to assess the sustainability of parallel software in practice, and also by defining parallel programming paradigms and scheduling policies intrinsically oriented to sustainability. As efficiency goals have led to the development of a significant number of offline auto-tuning libraries for performance optimization in the last decade, the current social and environmental situation urges us to consider additional goals, such as energy consumption, or wider-range, sustainability-related factors. The identification of ``best practices'' to promote sustainability at the software level could represent a crucial guideline for HPC developers.
The workshop’s objective is to foster the exchange of ideas on sustainable HPC algorithms, platforms, programming paradigms, and theoretical models, so as to, possibly, build a community of researchers working on these paramount topics.