This project serves as a performance benchmarking suite to measure and compare Agent-to-Agent (A2A) protocol communication and tool execution performance across multiple programming languages: Python, Go, Node.js (TypeScript), and Rust.

The benchmark runs an agent in each language that implements a tool to calculate the first $N$ Mersenne primes using the Lucas-Lehmer primality test. A central Master Coordinator Agent orchestrates runs from $N = 1$ to $N = 20$ (or higher) across the language agents to measure performance and capture execution time.

The suite consists of the following components:

• Master Coordinator Agent (Python): Orchestrates benchmarks by calling sub-agents via the A2A protocol and measuring execution times. Exposes a FastMCP interface on port 8100.
• Python Benchmark Agent: Calculates Mersenne primes using the Lucas-Lehmer test, exposed as an A2A app on port 8101.
• Go Benchmark Agent: Calculates Mersenne primes, built with the Go implementation of ADK, running on port 8102.
• Node.js Benchmark Agent: Written in TypeScript using @a2a-js/sdk, running on port 8103.
• Rust Benchmark Agent: Written in Rust using the axum web framework and num_bigint crate, running on port 8104.

Run the initialization script to set up your Google Cloud Project ID and Gemini API Key:

./init.sh

Source the environment variables required for running the benchmarks:

source set_env.sh

You can run each agent individually in separate terminals:

Runs the target Python prime calculation agent:

./bench-python.sh

Runs the Go benchmark agent:

./bench-go.sh

Runs the Node.js/TypeScript benchmark agent:

./bench-node.sh

Runs the Rust benchmark agent:

./bench-rust.sh

Runs the coordinator master agent that delegates tasks and benchmarks the other agents:

./bench-master.sh

To quickly verify that the Rust agent is functional and responding to A2A requests:

./test-rust.sh

The benchmark results can be plotted using:

python plot_primes.py

This generates prime_calculation_times.png. Existing benchmarking comparison plots are saved as:

• benchmark_performance.png (General comparison)
• benchmark_performance_15-20.png (Comparison for larger counts)