A Parallel And Resilient Frontend For High Performance Validation Suites

In any well-structured software project, a necessary step consists in validating results relatively to functional expectations. However, in the high-performance computing (HPC) context, this process can become cumbersome due to specific constraints such as scalability and/or specific job launchers. In this paper we present an original validation front-end taking advantage of HPC resources for HPC workloads. By adding an abstraction level between users and the batch manager, our tool JCHRONOSS, drastically reduces test-suite running time, while taking advantage of distributed resources available to HPC developers. We will first introduce validation work-flow challenges before presenting the architecture of our tool and its contribution to HPC validation suites. Eventually, we present results from real test-cases, demonstrating effective speed-up up to 25x compared to sequential validation time paving the way to more thorough validation of HPC applications.