A European I/O Trace Repository to Build Better I/O Systems & HPC Applications

This BoF aims to raise awareness on the European I/O trace repository. This initiative has been kicked-off at SC22 as a collaboration between several EuroHPC projects. Our purpose is to offer an open web portal, hosting I/O traces from as many HPC applications as possible. Traces will be openly available to the community: researchers, application developers or storage specialists. This repository will bring insight and knowledge by exposing the somehow speculated but never robustly quantified characteristics of the IO traffic from HPC applications. During the BoF, three aspects will be specifically discussed with the audience: Getting the traces: lowering the profiling entry cost on the user side is a critical step to the success of the project. The repository will therefore propose measurement receipts and pre-configured tools to ease capturing of I/O information. Hosting the traces: traces need to be stored in an interoperable format, augmented by metadata to describe the context of the trace collection. The BoF will present the metadata scheme and trace formats and steer a discussion with the audience on how to extend it in the future. Processing the traces: additionally to I/O traces, the repository will host a catalog of open-source analysis software and scripts from the community to extract value and information from the traces. The BoF will help to define future directions of the repository, it will offer an opportunity to involve other practitioners in the initiative and more generally to collect valuable feed-backs from our peers.