In this talk I will go over our experience with using Kokkos within the ADIOS2 framework for large-scale I/O storage and streaming. Our Kokkos backend enables the execution of large scale workflows consisting of multiple Kokkos applications running distributed across the globe exchanging data with each other, while the user can visualize the data in real time and optionally storing data to persistent storage. The Kokkos backend also allows us to compute derived quantities and metadata much more efficiently so that users can query what pieces of data they need based on properties that are not directly stored or streamed. The talk will cover our enabled features, performance and a wish list for what we would need from Kokkos to improve our library.
In this talk I will go over our experience with using Kokkos within the ADIOS2 framework for large-scale I/O storage and streaming. Our Kokkos backend enables the execution of large scale workflows consisting of multiple Kokkos applications running distributed across the globe exchanging data with each other, while the user can visualize the data in real time and optionally storing data to persistent storage. The Kokkos backend also allows us to compute derived quantities and metadata much more efficiently so that users can query what pieces of data they need based on properties that are not directly stored or streamed. The talk will cover our enabled features, performance and a wish list for what we would need from Kokkos to improve our library.
Slides: 2023_KUG_Gainaru.pdf