Functional Testing with STLs: A Step Towards Reliable RISC-V-based HPC Commodity Clusters.

The reliability of High-Performance Computing (HPC) systems is an essential concern due to their massive size and the complexity of their operation. Thus, functional tests have been extensively used to monitor HPC systems and use software routines to verify the software stack’s operation, mainly focusing on high-level abstraction features. However, the miniaturization of transistor technologies and the increment of computational resources (to face the performance and computation capabilities of HPC systems for the exascale generation) impose new reliability challenges that involve the development of clever testing strategies considering the underlying hardware characteristics. Interestingly, resorting to open-hardware architectures (such as RISC-V-based platforms) in the HPC domain offers a unique opportunity to effectively combine traditional HPC functional testing techniques with the adoption of effective fine-grain hardware testing solutions, such as those based on the Software-Based Self-Test (SBST) strategy. This work proposes the SBST strategy as an enhanced and complementary technique for functional testing of RISC-V platforms for HPC systems. The method provides fine-grain evaluations of the CPU cores, including quantitative information on the state of the CPU cores and the presence of faults. For the experiments, we resort to two RISC-V cores (RI5CY and ibex) to develop and verify the effectiveness of the SBST strategy. In total, we developed 11 STLs (SBST routines) showing that a considerable percentage of hardware faults (from about 82% and up to 90%) can be detected with minimal overhead, thus, allowing their use during empty time intervals or in combination with other in-field functional testing approaches for HPC clusters.