The road to a Universal Internet Machine
summary
Claim: building a widely distributed universal computer loses Turing universality unless consensus is available, but many real tasks don’t need consensus. Tools and results: reductions to consensus, state-machine replication, hardware primitives like compare-and-swap, RDMA (reducing 3f+1 to 2f+1), and the snapshot object implement a payment abstraction where only an owner invokes pay, yielding consensus number one for that payment problem. Implementations: prototype systems and algorithms (e.g. mu: 1 ms best case) exist and the payment and some Byzantine-tolerant ML tasks work without full consensus.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 2:10The quest for universality
- 4:37Historical context and models
- 9:01Computational equivalence
- 12:32Implementing objects
- 17:01Consensus and impossibility
- 22:50Circumventing impossibility
- 30:55Reducing consensus needs
- 35:47Review of research progress
- 40:20Practical case studies
- 48:28The role of programming languages
- 51:21Question and answer session
related talks
56mA sledgehammer to crack a nut: why blockchain is not (always) a good idea
38mTowards consensus on distributed consensus #bbuzz
31mProofless Consensus & Client-side Validation
43mWhat We Talk About When We Talk About Distributed Systems
48mFrom Byzantine Generals to Global Harmony, Modularly
48mDistributed Consensus Algorithms
51mEconomic Security of Distributed Systems
43mAll Together Now: An Introduction to Distributed Consensus