Scientific advances designed to address global challenges require researchers to have seamless access to data and computing and increasingly high performance computing. A certain disconnect has characterized the relationship between the HPC and data communities and this needs to be addressed in order to fully support today's data and compute intensive science.
For example, members of the ArtificiaI Intelligence/Machine Learning (AI/ML) community are starting to use the resources of HPC, as HPC systems include GPUs for acceleration. Reciprocally, AI/ML is an emerging method in science, the main consumer of HPC resources. Hybrid HPC/ML workflows enable combining ML tasks with traditional numerical simulations on HPC resources, for instance for data-intensive analysis, experiment optimization, and data mitigation when data values are missing. The use of HPC resources to run AI/ML presents new challenges for data, related to metadata and provenance, I/O, and reproducibility. Another example of topics requiring bridging the two communities is the increasing development of digital twins in environmental sciences that require bringing together diverse data from various sources and adapting that to HPC workflows.
This BoF session will continue the discussion started in RDA P18 session “Combining Data community strengths and High Performance Computing opportunities”, followed by BoF sessions “Leveraging Data community strengths and High Performance Computing opportunities”, and “Bridging the HPC/Data Divide” in IDW 2022, Supercomputing 2022 and International Supercomputing 2023 conferences. The purpose of these sessions has been to discuss topics bridging the divide between the communities, as well as challenges and solutions the communities share. Both communities have been asked for feedback on the divide, and the audiences have been actively engaged into the sessions.
The community feedback showed that both communities identify the divide and the need to share best practices, success stories and provide concrete solutions and data management techniques for HPC users. At the same time, it is evident that there are significant differences in terms of awareness of data management principles among the HPC community.
This BoF will wrap up the discussions and feedback collected from the previous sessions, and further elaborate efforts to bridge the gap and lessons learned from the work. A few topics identified based on the previous discussions will be considered more closely such as solutions in sensitive data management, and FAIR data in HPC environment. The expected outcome of this session is an increased understanding of the division between the communities, but also of the synergies and overlaps, opportunities for closer collaboration. The need for establishing an RDA working group or interest group will be discussed in view of overcoming the gaps and increasing understanding, common language, and ways of working between the communities.