High Performance Networking: How DoubleZero Enables IBRL
summary
DoubleZero builds a purpose‑built wide‑area network that currently spans an 8‑metro testnet and targets ~30 metros by end of 2025, choosing routes by measured latency to minimize hops (example: 22 hops on the public internet versus 5 hops on DoubleZero, saving ~12 ms end‑to‑end). The design uses modern switching ASIC features to keep queue watermarks written to the ledger, aims for <50% MLU in failure modes, and leverages kernel bypass in Anza and Firedancer to cut serialization and processing delays.
AI-written summary · about
chapters (8)
- 0:00Introduction to D0ero
- 0:50Defining network performance
- 2:21Sources of network latency
- 4:03High-performance WAN design
- 8:00D0ero architecture in action
- 12:44Network instrumentation
- 16:07D0ero vs. public internet
- 18:23Purpose-built networking
related talks
03mHow DoubleZero’s Physical Fiber Infrastructure Makes Blockchains 100x Faster Than Public Internet
03mWhy Crypto’s Next Breakthrough May Come From Hardware
14mHigh-throughput and Flexible Host Networking for Accelerated Computing
32mLive Q+A 2
17mMeta’s Network Journey to Enable AI
13mLive in the Express Lane
19mAI/ML Networking Challenges: The Fast and the Finicky!
15mCost-effective and Reliable Global Internet Peering with Programmable Switches