A benchmark of 23 web frameworks: the top 5 were Rust-based

#rust #ai #frameworks #web

🧪 Benchmark Methodology (4-Tier Architecture)

To simulate different real-world scenarios, our testing approach is divided into four robust tiers:

🌍 Tier 1: Fast Load (Baseline Performance)

Goal: Establish a baseline for raw Requests Per Second (RPS) and latency.
Load: 125 concurrent connections for 10 seconds.
Endpoints: /text, /json, /db-single, /html

⚡ Tier 2: Concurrency & Latency (High Traffic)

Goal: Evaluate behavior under sudden traffic spikes and measure P99 latency.
Load: 300 concurrent connections for 30 seconds.
Endpoints: /json

🧠 Tier 3: Resource Efficiency (Idle vs. Peak)

Goal: Identify CPU and RAM utilization of "greedy" vs. "efficient" architectures.
Method: Measure Docker stats during 0 traffic (Idle) vs. 200 concurrent connections (Peak).

🔥 Tier 4: Stress & Resilience (Memory Leak Detection)

Goal: Ensure frameworks don't crash or leak memory under prolonged stress.
Load: 500 concurrent connections for a sustained 2 minutes.
Endpoints: /json

🏅 The Efficiency Ranking Score (RPS/MB)

Our official ranking is determined by the Efficiency Score, which measures how many requests a framework can handle per megabyte of memory it consumes.

Formula: Efficiency = JSON RPS (Tier 1) / Peak RAM in MB (Tier 3)
Example: A framework handling 100,000 RPS with 20MB RAM scores 5,000. A framework handling 100,000 RPS with 200MB RAM scores 500.
The Disqualification Penalty: Any framework that drops connections, crashes, or returns timeout errors during the Tier 4 Stress Test is automatically disqualified and receives 0 Points. Speed is irrelevant if the server collapses under pressure!

💻 Minimum Hardware Requirements

To run these benchmarks locally without skewing the results due to resource starvation (especially during the 500-connection Tier 4 tests), we recommend the following minimum specifications:

CPU: 4 Cores / 8 Threads (8+ Cores recommended)
RAM: 8GB Minimum (16GB+ recommended)
OS: Linux, macOS, or Windows (WSL2)
Docker Engine: Ensure Docker is allocated at least 4 CPUs and 8GB of RAM.

📊 Official Tested Hardware

All benchmark results listed in this repository were run on the following environment to ensure consistency:

CPU: AMD Ryzen 7 5700U (8 Cores / 16 Threads)
RAM: 8GB
OS: Windows 11 (Docker with WSL2)

🗄️ Database Architecture

All frameworks connect to a dedicated PostgreSQL 15 container.

The database is initialized via the db-init scripts which pre-populate a table called world with thousands of random rows.
Frameworks must fetch the World record with ID = 1 for the /db-single endpoint. Connection pooling is strictly recommended.

The Top 10:

🏅 Official Efficiency Ranking

🏆 Rank	Framework	Language	Efficiency (RPS/MB)	JSON RPS	DB Single RPS	Avg Latency	Peak RAM	Status
🥇 1st	Actix-Web	Rust	6.778	128.654,31	12.366,9	2.34ms	18.98MiB	✅ Passed
🥈 2nd	Axum	Rust	4.580	100.620,33	18.033,16	2.61ms	21.97MiB	✅ Passed
🥉 3rd	Poem	Rust	4.346	98.795,48	12.405,22	2.68ms	22.73MiB	✅ Passed
4	Rullst 👑	Rust	4.344	90.485,67	16.186,7	3.14ms	20.83MiB	✅ Passed
5	Salvo	Rust	4.141	92.218,37	7.465,66	2.96ms	22.27MiB	✅ Passed
6	Go-Fiber	Go	4.032	99.803,89	16.026,64	2.68ms	24.75MiB	✅ Passed
7	Dioxus	Rust	3.448	87.638,05	9.431,99	2.79ms	25.42MiB	✅ Passed
8	Leptos	Rust	3.024	77.135,01	8.009	2.94ms	25.51MiB	✅ Passed
9	Go-Gin	Go	2.898	70.219,4	13.627,19	4.14ms	24.23MiB	✅ Passed
10	ASP.NET Core	C#	718	75.723,47	11.181,52	3.63ms	105.5MiB	✅ Passed