Crescent: Emulating Heterogeneous Production Network at Scale
Talkat 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24) · 2024-04-16..04-18·
summary
Crescent emulates production networks by running switch vendor images inside containers and links them with veth pairs and OVS VXLAN, trading full-network coverage for a canary testbed that includes all core devices and representative noncore devices. A graph-partitioning assignment minimizes cross-host links so partitioning reduces testbed bootup time by up to 20× versus random placement, and an incremental reconfiguration path makes live modifications up to 30× faster than rebuilding a testbed. Deployed for 3.5 years, Crescent measurably cuts change-induced incidents.
AI-written summary · about
chapters (7)
- 0:00Crescent introduction and goals
- 2:51Incident example: network change risks
- 8:30Three years of network incidents
- 9:55Network symmetry and system design
- 10:58Crescent system implementation
- 13:52Evaluation and production impact
- 16:24Summary and future directions
related talks
29mLiving on the Edge with DCN Networking
19mDisaggregating Stateful Network Functions
01h 21mIntro + Deep Dive: Cloud Native Network Function (CNF) Testbed
44mHow we designed an OpenStack Public Cloud at Scale with Fabric Networking
18mCStar Gateway: Augmenting Public Cloud Infrastructure for Heterogeneous Network...
38mHow can a network be "virtual"?
21mMirrorNet: High-fidelity and Scalable Network Emulation for Software-defined WAN
17mMeta’s Network Journey to Enable AI
