April 22-25, 2019

Theme: Entering an Era of Extreme Heterogeneity

Over the last decade, the field of high-performance computing has seen extreme-scale high performance computing systems become increasingly more heterogeneous. Various types of accelerators have been employed to achieve greater levels of compute performance. Adapting applications to use these accelerators effectively has involved a significant amount of effort, not only in code modifications, but also in developing the programming environment. While several applications have been able to take advantage of heterogeneous computing systems, many still have not. Several industry trends seem to indicate that large-scale homogeneous platforms, which contain a single type of compute engine with a relatively straightforward memory hierarchy, will soon be a thing of the past. The diversity that has appeared in computing technology will likely continue to include more specialized hardware components as power and energy concerns motivate the need for more effective use of transistors. The convergence of memory and storage technologies is increasing the complexity and depth of the memory hierarchy. It is becoming more difficult to map programming abstractions to the hardware effectively, and programming models no longer reflect the important costs associated with critical resources. Hardware heterogeneity leads to performance heterogeneity and optimizing in the presence of non-determinism is extremely challenging. Applications are also evolving to include a more diverse set of capabilities. Traditional high-performance computing applications are expanding to include more functionality, such as support for in-situ analytics, uncertainty quantification, and the ability to compose components to build more complex workflows. New applications are being developed in higher-level languages and frameworks, creating scalability and integration challenges. Container technology is being leveraged to overcome the shortcomings of a single, one-size-fits-all software environment. The limitations of large batch-scheduled space-shared computing systems in supporting more diverse sets of applications are becoming more evident. The expansion of heterogeneity in multiple dimensions - compute, memory, storage, systems, software, applications, and usage models - characterizes Extreme Heterogeneity. The five sessions for the Salishan 2019 conference will explore the challenges and ramifications of this coming era of Extreme Heterogeneity.