Scaling a Web App: From 10 Users to 10,000

At around 200 to 300 people browsing at once, one server starts sweating. Pages took four seconds to load. The fix is a load balancer, which is a traffic cop that spreads incoming requests across two or more servers, so no machine gets swamped.

That creates a new problem: sessions. A session is how a server remembers you're logged in. If your booking lands on server A and your next click lands on server B, you look like a stranger again. Store that memory somewhere shared instead, like Redis (a storage service that keeps small pieces of data in memory for speed). Then any server can pick up any request.

Auto scaling belongs here too. It adds servers when traffic spikes and shuts them off overnight, so you pay for the lunch rush and not the quiet hours.

The database is what actually breaks

This part surprised him. The database (the program that stores and fetches your data) was fine at 1,000 users and choking at 4,000, even with four web servers. Three fixes, in order of payoff:

• Connection pooling. Every query needs a connection, like a taxi at a rank. Ten servers with 20 connections each is 200 taxis, and the database refuses to serve that many. Pooling keeps a small reusable set and shares it.
• Indexes. An index is like the index at the back of a book. His "find an open slot" search took 4 seconds across 80,000 bookings. One index cut it to 40 milliseconds.
• Caching. A cache is a shelf of answers you keep nearby. The shop list and opening hours change once a month, so they shouldn't be rebuilt for every visitor. A $15-a-month cache service absorbed most of his read traffic.

What this means for you

• Don't buy the biggest server on day one. Start small, watch it, upgrade when it hurts.
• Check two numbers weekly: your database's CPU percentage and your slowest query. Every cloud dashboard shows both.
• Set a billing alert at twice your normal spend. Scaling problems often arrive as a surprise invoice first.
• Never switch on auto scaling without connection pooling. More servers can make a crowded database worse.

For the record, his bill at 10,000 users was about $340 a month. Roughly $120 of that went to the database layer, not the servers.

Key Takeaways

• One small server handles your first few hundred users, so start cheap.
• A load balancer lifts you past that, but only if logins are stored in a shared place.
• Slow queries, not slow hardware, cause most slowdowns after 1,000 users.
• Cache anything that barely changes, and index anything you search by.

Frequently Asked Questions