Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Learn about a novel CPU-NIC interface design that addresses performance bottlenecks in network packet processing through this 18-minute conference presentation from OSDI '25. Explore how traditional per-core receive rings create memory bandwidth pressure due to large I/O working sets that exceed LLC capacity, and understand the limitations of existing solutions like shRing systems under imbalanced loads. Discover the rxBisect approach, which decouples memory allocation and packet delivery by replacing single Rx rings with two separate ring structures, enabling efficient buffer sharing between cores during load imbalances. Examine the implementation details through software emulation and analyze performance improvements of up to 20% over state-of-the-art solutions and 37% over traditional per-core approaches, presented by researchers from EPFL, NVIDIA, Tel Aviv University, and Technion.
