HACO: a highly available, causally ordered, distributed group chat

Politecnico di Milano - Distributed Systems (A.Y. 2023/2024)

This project implements a fully distributed group chat application, where users can create chat rooms, post messages, and participate in conversations with causal message delivery, all without relying on any centralized server.

Project Description

The application supports:

• Room Management: Users can create and delete chat rooms with a fixed set of participants specified at room creation.
• Fully Distributed Architecture: Communication between users (peers) occurs without a central server.
• High Availability: Users can continue using the chat even during temporary network disconnections.
• Causal Delivery: Messages within each room are delivered in causal order to ensure consistency in conversation history.

Key Features

• Distributed Messaging: All communication is done in a peer-to-peer (P2P) manner.
• LAN Connectivity: Peers are connected over LAN.
• Reliable Clients: Clients can join and leave the network at any time. Failures or partitions in the network are tolerated.
• Java Serialization: Messages are serialized using Java's native serialization.
• UDP Multicast/Unicast: Communication between peers uses unicast and multicast.

Software Architecture

P2P Model - Gnutella-like Architecture

• Fully Distributed: There is no central authority or supernode in the network. Peers discover each other dynamically.
• Dynamic Topology: No fixed topology is required. New peers join the network via a broadcast-based discovery mechanism.
• Ping-Pong Protocol: When a peer joins the network, it sends a PING message to discover other peers. Peers respond with a PONG, sharing their information for future communication.

Distributed Protocol Overview

• Messaging: Peers exchange messages via UDP multicast. A dedicated multicast listener runs on a separate thread for each room.
• Acknowledgements: Communication is made reliable using an acknowledgment system for message delivery.
• Vector Clocks: Causal delivery of messages is ensured using vector clocks to maintain message ordering within each chat room.

Technologies Used

• Language: Java
• Networking: UDP sockets (unicast and multicast)
• Serialization: Java Object Serialization
• Multithreading: Each room runs a separate listener for handling multicast messages.

Assumptions and Constraints

• Reliable Clients: Clients are assumed to close the application in expected ways, ensuring that they do not leave the network abruptly.
• No Centralized Authority: The application is fully distributed, with no supernode or server.
• Network Failures: The system is resilient to network partitions, allowing peers to reconnect and continue messaging.