How Nginx Was Able to Support 1 Million Concurrent Connections on a Single Server ✨

#62: Break Into Nginx Architecture (4 Minutes)

Neo Kim

Dec 04, 2024

200

Get the powerful template to approach system design for FREE on newsletter sign-up:

This post outlines Nginx architecture. You will find references at the bottom of this page if you want to go deeper.

Share this post & I'll send you some rewards for the referrals.

Note: This post is based on my research and may differ from real-world implementation.

Once upon a time, there was a tech startup.

They offered invoice generation services.

Yet they had only a few users.

So they ran their site on a single web server - Apache HTTP.

Life was good.

But one day they started to receive massive traffic.

Yet their server had only limited capacity.

So they set up more servers.

Although it temporarily solved their scalability issues, there were newer problems.

Here are some of them:

1. Resource Usage

A connection to the web server needs a separate thread or process.

Also consumes memory and CPU time until it’s released.

So computing resources get wasted.

2. Performance

A thread must wait until a blocking task is complete.

Besides context switch increases with many threads.

Think of the context switch as switching a CPU between different threads.

So performance is bad.

Onward.

Eraser - Sponsor

Eraser is an AI co-pilot for technical design that can help your team deliver accurate, consistent designs faster. Create beautiful technical diagrams in seconds from natural language prompts or code prompts (Terraform or SQL).

How Does Nginx Work

They wanted to reduce costs and ditch the complexity of many servers.

So they moved to Nginx.

And here’s how it offers extreme scalability:

1. Parallelism

Running many tasks at the same time is called parallelism.

They built Nginx with the master-worker model:

The master reads the configuration file and creates new workers.
While the worker handles client connections.

A simple approach is to give each connection a separate thread or process.

Yet it will result in context switches, thread trashing, and memory usage.

Imagine thread trashing as simply switching between threads instead of doing actual work.

So they run each worker as a separate process. And assign a separate CPU to each worker. Put simply, the number of workers equals the number of CPUs.

It gives better performance.

Ready for the best part?

2. Concurrency

Managing many tasks at the same time is called concurrency. (Yet all tasks don't have to run at the same time.)

A Worker Processing Requests Using the Event Loop

They run the worker as a single-threaded process:

The client sends a request to the worker.
The request gets added to its event queue.
The worker processes the request using the event loop.

While event loop is asynchronous, event-driven, and non-blocking.

Think of the event loop as the chef in a busy kitchen - they prepare many orders by doing 1 task at a time. And switch between orders when ready.

A Worker Delegating Blocking Tasks to the Thread Pool — A Worker Delegating a Blocking Task to the Thread Pool

A CPU-intensive task will keep the event loop busy. (Also block the worker from accepting new requests.)

So they offload blocking tasks to the thread pool.

Think of the thread pool as a group of threads created at Nginx startup. And they're shared between workers.

Workflow of Processing a Blocking Request

Here’s how it works:

A worker delegates a blocking task to the thread pool.
The worker then handles other requests using the event loop.
The thread pool tells the worker once the blocking task is complete.
The worker responds to the client.

This approach keeps the event loop responsive.

Ready for the next technique?

3. Scalability

A simple way to scale is by caching responses.

Yet it’s inefficient to cache the same data across different workers.

So they set up shared memory.

Workers Accessing the Same Data via Shared Memory

It let them:

Share cache data across workers.
Store session data and route requests to a specific backend server.
Rate limit by tracking requests.

Besides it reduces the memory usage of Nginx.

While workers use a lock (mutex) to access shared memory.

A single Nginx server can handle 1 million concurrent connections.

Yet it isn’t a silver bullet to scalability.

So choose the right tool for your needs.

👋 PS - Are you unhappy at your current job?

And preparing for system design interviews to get your dream job can be stressful.

Don't worry, I'm working on content to help you pass the system design interview. I'll make it easier - you spend only a few minutes each week to go from 0 to 1. Yet paid subscription fees will be higher than current pledge fees.

So pledge now to get access at a lower price.

“This newsletter is an amazing resource for learning system design.” Alex

Subscribe to get simplified case studies delivered straight to your inbox:

Neo’s Recommendations

Product for Engineers: Actionable advice to improve your product skills as a software engineer.
The Developing Dev: Mentorship from a staff engineer at Instagram.