Skip to content

Benchmarks

Zero is built for performance. Here's how it compares to other frameworks.

Test Environment

  • Hardware: 13th Gen Intel Core i9-13900HK @ 5.40GHz
  • Cores: 14 cores, 20 threads
  • RAM: 32GB
  • OS: Ubuntu 24.04 LTS
  • Containerization: Docker

Test Scenarios

Scenario 1: Simple Hello World

A simple endpoint returning a constant response. Tests baseline throughput and latency.

Endpoint: /hello - Returns "hello world"

Scenario 2: Redis Save

Tests I/O performance - each request saves an Order object to Redis.

Endpoint: /order - Saves order data to Redis

Results

Framework hello world (req/s) 99% latency (ms) redis save (req/s) 99% latency (ms)
zero (tcp) 100,752 2.33 35,813 13.48
sanic 58,811 4.43 23,623 9.22
zero (async) 41,092 4.41 23,996 8.64
blacksheep 38,026 8.41 16,324 13.54
aiohttp 33,168 11.89 17,959 12.76
zero (sync) 27,571 6.65 10,269 23.71
aiozmq 25,174 6.13 8,850 10.19
fastapi 19,683 9.09 12,776 16.28

When Zero is Best

  • Building RPC microservices
  • High throughput is needed (10k+ req/s)
  • Low latency is critical
  • Simple request-response patterns
  • Optimizing infrastructure costs

When Other Frameworks Might Be Better

  • Complex routing requirements
  • REST API conventions
  • HTML templating

Running Benchmarks

Zero includes benchmark tools for testing against other frameworks.

See benchmarks/dockerize for setup instructions.

Performance Tips

To maximize Zero performance:

  1. Use TCP protocol instead of ZeroMQ

    from zero.protocols.tcp import TCPServer
app = ZeroServer(protocol=TCPServer)

  2. Use async for I/O operations

    @app.register_rpc
async def fetch_data(url: str) -> str:
    # Async I/O here

  3. Reuse client connections

    client = ZeroClient("localhost", 5559)
# Reuse for multiple calls

  4. Avoid blocking operations in RPC handlers

  5. Use appropriate worker count

    app = ZeroServer(workers=None)  # Default: CPU cores

  6. Keep messages small - Reduces serialization overhead

Further Reading