shardcast 0.3.2

A package for distributing large files via HTTP P2P network

Shardcast

A Python package for distributing large files via an HTTP-based tree-topology network

Overview

Shardcast is designed to distribute large binary files through a multi-tier network, making it efficient to transfer large files to many clients:

1. Origin Server: The root node that shards a large file and serves the shards via HTTP
2. Middle Nodes: Intermediate servers that download shards from upstream servers and re-serve them
3. Client Nodes: End nodes that download and reassemble shards into the original file

Features

• Automatically shards large files into configurable chunks (default: 50MB)
• Versioned distribution with auto-cleanup of old versions
• SHA-256 integrity verification for reassembled files
• Dynamic server performance tracking for optimal downloads
• Concurrent downloads with automatic retries
• Support for multiple distribution layers
• Simple API for broadcasting files

Installation

# Install from source
git clone https://github.com/PrimeIntellect-ai/shardcast.git
cd shardcast
pip install -e .

Usage

Origin Server

Run as a standalone server:

# Start an origin server on port 8000
shardcast-origin --data-dir ./data --port 8000

Use as a library:

import shardcast

# Initialize the package
shardcast.initialize(data_dir="./data", port=8000)

# Broadcast a file
version = shardcast.broadcast("/path/to/large_file.bin")
print(f"File broadcast as version {version}")

# Shut down when done
shardcast.shutdown()

Middle Node

# Start a middle node that connects to an origin server
shardcast-middle --upstream 192.168.1.100 --data-dir ./middle_data --port 8001

# Connect to multiple upstream servers (comma-separated)
shardcast-middle --upstream 192.168.1.100,192.168.1.101 --data-dir ./middle_data --port 8001

# Using the IP_ADDR_LIST environment variable instead of --upstream
export IP_ADDR_LIST="192.168.1.100 192.168.1.101"
# or in bash array format
export IP_ADDR_LIST=("192.168.1.100" "192.168.1.101")
shardcast-middle --data-dir ./middle_data --port 8001

Client Node

# List available versions
shardcast-client --servers 192.168.1.100,192.168.1.101 --list

# Download a specific version
shardcast-client --servers 192.168.1.100,192.168.1.101 --version v1 --output-file ./downloaded_file.bin

# Using the IP_ADDR_LIST environment variable instead of --servers
export IP_ADDR_LIST="192.168.1.100 192.168.1.101"
# or in bash array format
export IP_ADDR_LIST=("192.168.1.100" "192.168.1.101")
shardcast-client --list

Configuration

Key constants are defined in shardcast/constants.py:

• SHARD_SIZE: Size of each shard in bytes (default: 50MB)
• MAX_DISTRIBUTION_FOLDERS: Maximum number of version folders to keep (default: 15)
• HTTP_PORT: Default HTTP port for servers (default: 8000)
• RETRY_ATTEMPTS: Number of retry attempts for failed downloads (default: 5)
• MAX_CONCURRENT_DOWNLOADS: Number of concurrent download threads (default: 10)

Architecture

• File Sharding: The origin server splits files into shards named shard_001.bin, shard_002.bin, etc.
• Distribution: Shards are served via HTTP from the origin server and middle nodes.
• Folder Versioning: Each broadcast creates a new folder (e.g., v1, v2), with a maximum of 15 folders.
• Discovery: A distribution.txt file lists active shard folders and their blake3 checksums.
• Download Optimization: Clients download concurrently and prefer faster middle nodes based on runtime performance.
• Integrity: Clients verify the reassembled file using the blake3 checksum from distribution.txt.

cat distribution.txt
> v1: 4d1d960b53356285f45ea2e27c89a1a11d10a9601d3ba2a90851f9f227dd9295