As computing systems become more powerful, traditional post-processing methodologies need to be replaced by alternative in-situ processing in which the data are visualised and analysed as they are generated. This in-situ visualization project represents the continuation of a longstanding collaboration between Intel, TACC, Kitware and the Cosmos Intel Parallel Computing Centre (IPCC), hosted in the Cambridge Faculty of Mathematics. Enhanced in-situ visualization capababilities will be publicly released in the open-source OSPRay and Catalyst libraries and will be incorporated in future releases of the Paraview visualization application. Our goal is also to apply this programming model to exemplar community codes allowing them to be used across diverse accelerator architectures.
Specific objectives will include extending robust in-situ capabilities to composite AMR distributed data using the Paraview/Catalyst/OSPRay libraries. The key initial demonstrator is GRChombo, a community numerical relativity code, after which we will assist with the introduction of in situ capabilities to other ExCALIBUR codes. We will also develop pioneering work on efficient paradigms for shared-memory offload to multiple accelerators. An online training course in-situ visualization will be offered to disseminate these advances to other ExCALIBUR projects and the wider UK HPC community.
