Abstract

With the recent arrival of exascale HPC systems and the expectation of more to come, there is a pressing need to develop new methodologies in order to exploit their potential. Until recently the dominant approach to visualisation and analysis of outputs from scientific simulations has been post-hoc processing where data saved to disks during the simulation is loaded back into memory and processed accordingly. Unfortunately, the I/O bandwidth and capacity of HPC file systems have not kept up with the increase in processing power meaning that it is becoming more impractical to continue the current post-processing paradigm for the largest simulation datasets. The alternative approach of in-situ processing, where data is processed during the simulation, is becoming more widespread but there is still significant development required in order to bring these in-situ frameworks up to scratch with that of post-processing ones. In collaboration with Intel, we have been developing an in-situ visualization capability for the Adaptive Mesh Refinement (AMR) numerical relativity code GRChombo using ParaView Catalyst. ParaView uses Intel’s OSP Ray library in order to perform ray-tracing for volume rendering. Currently, OSP Ray’s support for AMR datasets that are MPI -distributed is limited which means that workarounds need to be applied in order to use it in-situ.

In this talk, I will explain the process of instrumenting a code with ParaView Catalyst that will allow it to perform in-situ processing and share my experiences of doing so with GRChombo. I will also discuss the recent work with Intel to improve the support for distributed AMR data in OSP Ray.

Hybrid Zoom link: https://zoom.us/j/94539775650?pwd=TjRGem5XMXV2YnBYVmpBRzhpNzNGZz09

Zoom Meeting ID: 945 3977 5650 Zoom Passcode: 496240