The past decade saw renewed interest in decentralized systems, fueled by their promises to harden vital services in our society against intentional and unintentional corruption, and to equalize access to these services. A key enabler of modern decentralized applications are blockchains, which are fault-tolerant distributed systems operated by trust-minimized entities across the world. Blockchains are inherently network systems, but existing designs critically overlook the dynamics of real-world networks such as variability and congestion. In this talk, I will explain how this mismatch leads to performance bottlenecks, security vulnerabilities, and operational challenges that are hindering the goal of decentralization for blockchains. I will then discuss two techniques to close this gap: (1) DispersedLedger, a network-aware decomposition of Byzantine-fault-tolerant state machine replication (BFT) protocols, which enables blockchains to robustly achieve high throughput and low latency in variable networks like the Internet; (2) a unified framework to analyze and secure longest-chain BFT protocols under bandwidth constraints. These techniques use tools from distributed systems, security, and computer networks, and have wide-ranging implications for the entire blockchain system stack.
Lei Yang is a PhD candidate at MIT advised by Prof. Mohammad Alizadeh. His research focuses on the intersection of networks and blockchains, using tools from distributed systems, computer networks, and security. A theme of his works is to examine the explicit and implicit assumptions that existing blockchain systems make of the network, and design novel techniques to bridge the gap between the assumptions and the reality. His works are backed by high-performance system implementations and end-to-end evaluations in real-world scenarios.
- Electrical and Computer Engineering