Paper Title: Network Stack that Allocates Resources Independently in Each Layer

Paper Abstract: The packet processing pipelines in today’s typical network stacks are dedicated, tightly integrated, and static. These characteristics preclude them from fully exploiting the capabilities of modern hardware. NetChannel, a new disaggregated network stack architecture for μs-scale applications running atop Terabit Ethernet, enables independent scaling and scheduling of resources allocated to each layer. A demonstration using an end-to-end NetChannel realization within the Linux network stack shows that NetChannel enables new operating points, such as: (1) enabling a single application thread to saturate multi-hundred gigabit access link bandwidth; (2) enabling near-linear scalability for small message processing with the number of cores, independent of the number of application threads; and, (3) enabling isolation of latency-sensitive applications, allowing them to maintain μs-scale tail latency even when competing with throughput-bound applications operating at near-line rate. NetChannel can be fully offloaded to a SmartNIC.

Paper Author: Qizhe Cai, PhD Student, Cornell

Author Bio: Qizhe Cai is a doctoral student in the Department of Computer Science at Cornell University. He got his MS in computer science at Princeton. His current research interests are designing and building performance models, network stacks, network protocols, and network hardware for Terabit Ethernet. He has published five articles on network software in major networking conference proceedings such as ACM SIGCOMM and USENIX NSDI. He has held internships at Google and Zazzle.