Calendar

Understanding and improving the performance of applications on NUMA systems

Modern multicore systems are based on a Non-Uniform Memory Access (NUMA) design. Efficiently exploiting such architectures is notoriously complex. In this talk we will present different works we did during these last years to help application and system designers leveraging the power of these architectures. In particular, we will present some profiling tools that we used to detect inefficiencies and some algorithms/heuristics that we implemented within the Linux kernel to improve the performance of NUMA systems.