Supercharging Packet-level Network Simulation of Large Model Training via Memoization...
summary
Wormhole implements a C++ ns-3 kernel that uses network partitioning, state memoization and rate-based steady-state detection to skip repetitive packet events and replay cached unsteady transitions via a Flow Contention Graph. It reports 227–745× speedups for GPT workloads (510× for MoE), >1,000× peak when paired with Unison, average FCT error <1%, steady-state skip >99% of events, and database storage below megabytes, demonstrating large practical acceleration with bounded error.
AI-written summary · about
chapters (11)
- 0:00Wormhole introduction
- 3:30Limitations of simulations
- 4:35Steady states and flow patterns
- 6:43Wormhole architecture
- 8:11Simulation memoization
- 9:36Steady state fast-forwarding
- 11:21Wormhole acceleration example
- 13:10Implementation details
- 13:30Evaluation results
- 15:36Summary and future work
- 16:18Q&A
related talks
16mAccelerating Distributed Graph Learning by Using Collaborative In-Network...
14mBrain-on-Switch: Towards Advanced Intelligent Network Data Plane via NN-Driven Traffic...
16mAutomatic Parallelization of Software Network Functions
20mNetKernel: Making Network Stack Part of the Virtualized Infrastructure
12mDebugging Transient Faults in Data Centers using Synchronized Network-wide Packet
21mCo-opting Linux Processes for High-Performance Network Simulation
16mState-Compute Replication: Parallelizing High-Speed Stateful Packet Processing
32mSnabb Switch simple and fast packet networking
