In a previous post, we show that State Machine Replication for any $f<n$ failures is possible in the synchronous model when the adversary can only cause parties to crash. In this post, we show that omission failures are more challenging. Implementing SMR requires at most $f<n/2$ omission failures. [Read More]
Primary-Backup State Machine Replication for Crash Failures
We continue our series of posts on State Machine Replication (SMR). In this post we discuss the most simple form of SMR: Primary-Backup for crash failures. We will assume synchronous communication. For simplicity, we will consider the case with two replicas, out of which one can crash. Recall that when a party crashes, it irrevocably terminates. [Read More]
A Payment Channel is a two person BFS-SMR system
This posts views payment channels as essentially a two person BFS-SMR system along with a carefully implemented mechanism for safe termination (channel closing) under assumptions of synchrony. [Read More]
Flavours of State Machine Replication
State Machine Replication is a fundamental approach in distributed computing for building fault tolerant systems. This post is a followup to our basic post on Fault Tolerant State Machine Replication. [Read More]
Flavours of Broadcast
What is the difference between broadcast, crusader broadcast, gradecast, weak broadcast, detectable broadcast, and broadcast with abort? This post is a follow up to our basic post on: What is Broadcast? [Read More]
Consensus for State Machine Replication
We introduced definitions for consensus, Byzantine Broadcast (BB) and Byzantine Agreement (BA), in an earlier post. In this post, we will discuss how consensus protocols are used in State Machine Replication (SMR). We will compare and contrast this setting to that of traditional BB and BA. A follow up post discusses the reductions from one abstraction to the other in the omission failure model. [Read More]
Flavours of Partial Synchrony
This is a follow up post to the post on Synchrony, Asynchrony and Partial synchrony. The partial synchrony model of DLS88 comes in two flavours: GST and Unknown Latency. In this post we discuss: [Read More]
Dont Trust. Verify. and Checkpoint?
Imagine that that Aliens land on earth with a new superfast SHA256 machine. Imagine this machine always gives them more than 51% of the current world Bitcoin hash power (but not enough hash power to completely break SHA256). Suppose they decide to build a chain from the Bitcoin Genesis block that is longer than any other chain on earth and put only empty blocks on it. Could they erase all... [Read More]
The Dolev and Reischuk Lower Bound: Does Agreement need Quadratic Messages?
How scalable is Byzantine agreement? Specifically, does solving agreement require the non-faulty parties to send a quadratic number of messages (in the number of potential faults)? In this post, we highlight the Dolev and Reischuk lower bound from 1982 that addresses this fundamental question. [Read More]
Byzantine Agreement is Impossible for $n \leq 3 f$ if the Adversary can Simulate
The Fischer, Lynch, and Merritt, 1985 lower bound states that Byzantine agreement is impossible if the adversary controls $f>n/3$ parties. It is well known that this lower bound does not hold if there is a PKI setup. [Read More]
The Trusted Setup Phase
co-authored with Avishay Yanai [Read More]
Do Bitcoin and Ethereum have any trusted setup assumptions?
What is Consensus?
We all broadly understand “consensus” as the notion of different parties agreeing with each other. In distributed computing, Consensus is one of the core functionalities. In this post, we define the consensus problem and discuss some variants and their differences. [Read More]
Byzantine Agreement is impossible for $n \leq 3 f$ under partial synchrony
Lower bounds in distributed computing are very helpful. Obviously, they prevent you from wasting time trying to do impossible things :-). Even more importantly, understanding them well often helps in finding ways to focus on what is optimally possible or ways to circumvent them by altering the assumptions or the problem formulation. [Read More]
What is the difference between PBFT, Tendermint, SBFT and HotStuff ?
In this post I will try to compare four of my favorite protocols for Byzantine Fault Tolerant (BFT) State Machine Replication (SMR): [Read More]
The threshold adversary
In addition to limiting the adversary via a communication model synchrony, asynchrony, or partial synchrony, we need some way to limit the power of the adversary to corrupt parties. [Read More]
The power of the adversary
After we fix the communication model, synchrony, asynchrony, or partial synchrony, and a threshold adversary we still have 5 important modeling decisions about the adversary power: [Read More]
Synchrony, Asynchrony and Partial synchrony
In the standard distributed computing model, the communication uncertainty is captured by an adversary that can control the message delays. The communication model defines the limits to the power of the adversary to delay messages. [Read More]