Crescent: Emulating Heterogeneous Production Networks at Scale

Explore the design, implementation, evaluation, and deployment of Crescent, ByteDance's network emulation platform, in this conference talk from NSDI '24. Learn how Crescent prevents change-induced network incidents by running switch vendor images inside containers for high-fidelity emulation. Discover the unique approach to scaling up the emulator, including exploiting data center network architecture symmetry and modularity to balance coverage and resource cost. Examine the node-to-host assignment process, formulated as a graph partitioning problem, which reduces testbed bootup time by up to 20 times compared to random partitioning. Understand the incremental approach developed to modify the emulated network on the fly, achieving 30 times faster modifications than creating new testbeds of the same scale. Gain insights into Crescent's active use over three and a half years, its significant impact on reducing change-induced network incidents, and its success in various other use cases. Benefit from critical lessons learned during Crescent's deployment in large-scale network environments.

NSDI '24 - Crescent: Emulating Heterogeneous Production Network at Scale