Webinar: HDF5 Application Tuning: There is more than one way to skin a cat(fish), part 1 supporting materials
On October 2, Gerd Heber presented a webinar, HDF5 Application Tuning: There is more than one way to skin a cat(fish), part 1. This post contains the supporting material for that webinar.
With HDF5, we don’t get too many complaints about the lack of options.
It’s more likely that users are confused about too many options to achieve seemingly the same end. ‘Seemingly,’ because the performance of different approaches can be all over the map. In this presentation, we show a simple but practically relevant example of such drastic performance differences (a factor of four between the best and worst performers!). Since it is relatively simple, we can offer a systematic step-by-step analysis and diagnosis of why that is the case.
This presentation continues our series dedicated to troubleshooting performance problems, the effective use of diagnostic tools, and how to make the most of HDF5 and its unique capabilities. It is part one of two presentations. In part one, we ignore additional (!) complications introduced by the use of parallel file systems, to which we will return in part two of the presentation.
Github Repo for code samples, etc.