The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|Theory and Practice in HPC: Modeling, Programming, and Networking|
(Presentation - presented in Xi'an, China, Oct. 2016)
Keynote talk at HPC China 2016
AbstractWe advocate the usage of mathematical models and abstractions in practical high-performance computing. For this, we show a series of examples and use-cases where the abstractions introduced by performance models can lead to clearer pictures of the core problems and often provide non-obvious insights. We start with models of parallel algorithms leading to close-to-optimal practical implementations. We continue our tour with distributed-memory programming models that provide various abstractions to application developers. A short digression on how to measure parallel systems shows common pitfalls of practical performance modeling. Application performance models based on such accurate measurements support insight into the resource consumption and scalability of parallel programs on particular architectures. We close with a demonstration of how mathematical models can be used to derive practical network topologies and routing algorithms. In each of these areas, we demonstrate newest developments but also point to open problems. All these examples testify to the value of modeling in practical high-performance computing. We assume that a broader use of these techniques and the development of a solid theory for parallel performance will lead to deep insights at many fronts.