Poster: exascale in-situ visualization using raytracing

The traditional visualization workflow stores simulation data to disk, followed by an independent analysis and visualization step as a post-process. This post-processing model compute-store-analyze will not scale on exascale machines. Already at terascale, I/O represented a significant bottleneck, and this problem will become a major bottleneck in the exascale era [1]. The access speed of disk drives is several orders of magnitude smaller than that of cpu-to-memory accesses. Thus the need to avoid I/O to disk, in favor of insitu visualization: visualize while the simulation is running, gaining direct access to the data using the resources allocated by the simulation code. In this model, the simulation advances in time using the normal iterative processing with regular hand-offs to service visualization requests. In-situ visualization is an attractive strategy to mitigate the effects of the I/O bottleneck.

Additionally, we consider data movement too inefficient in an exascale system and propose to do all computations locally where the data is created. This choice however precludes most means of load-balancing specifically for visualization. In this work, we investigate the imbalances involved in such an approach as well as its scalability and impact on the system.