HiPC 2016 Workshop 3: Workshop on High Performance Computing and Big Data in Molecular Engineering (HBME)

Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. The relevance of molecular modelling and simulation for engineering applications has greatly increased within the last decade due to the shift in the accessible length and time scales facilitated by massively parallel highperformance computing. Reaching quantitative agreement with the available experimental data, and predicting thermophysical properties where experimental data are absent, molecular engineering constitutes a promising approach for optimizing engineering processes. In particular, mechanical engineering (tribology, rheology, fatigue, machining at the molecular level) as well as chemical and process engineering (process design, thermodynamic properties, phase transitions, and interfacial phenomena) contribute to these developments significantly and are thereby qualitatively transformed, reinforcing the general tendency towards the virtualization of manufacturing and production processes. Prompted by the emergence of molecular engineering as a discipline which employs and develops approaches within HPC as well as mathematical modelling, simulation, and optimization, the Workshop on HPC and Big Data in Molecular Engineering (HBME 2016) at HiPC 2016 in Hyderabad, India, will address related challenges to high performance computing. Current and upcoming supercomputers consist of hundred-thousands to millions of cores. For simulations that are carried out over a substantial amount of time on huge core counts, specific challenges emerge. The resilience of the code with respect to single-node failures becomes an issue, potentially requiring redundant checkpointing, while the induced communication and I/O effort needs to be limited. Load balancing approaches need to incorporate strategies for processes with a fluctuating performance. Furthermore, big-data aspects and issues related to problems which inherently decompose into tasks with a significant, but non-trivial internal concurrency become relevant. This includes model parameterization, optimization, and validation as well as uncertainty quantification for simulation results, the evaluation of the soundness and internal consistency of the provided body of experimental data, as well as sampling, visualization, and analysis of problems with highdimensional order parameter spaces.