Course contentsShow
← System Design
Lesson 296 of 1,9198. Database Performance OptimizationFree lesson

Write Amplification Costs

Understand how denormalization multiplies write operations and increases database load.

Write Amplification Costs

What you'll learn: How denormalization forces you to update data in multiple places, multiplying write operations and database load.

The Hidden Cost of Denormalization

When you denormalize data (copy the same information to multiple tables or documents), you gain faster reads—but you pay a steep price on writes. Every logical update now requires multiple physical write operations.

How Write Amplification Happens

Imagine you store a user's name in a normalized system: it lives in one place (the users table). When the user changes their name, you write once.

Now imagine you've denormalized by copying that name into every order, comment, and post they made. When they change their name, you must update:

  • The users table
  • Hundreds of orders records
  • Thousands of comments records
  • Dozens of posts records

One logical write became thousands of physical writes. This is write amplification.

Real Costs

Write amplification impacts your system in concrete ways:

  • Database load: More UPDATE queries mean higher CPU, I/O, and lock contention
  • Write latency: Updates take longer because multiple operations must complete
  • Transaction complexity: Keeping denormalized data consistent requires larger transactions or eventual consistency patterns
  • Scaling costs: Write-heavy workloads are harder and more expensive to scale than read-heavy ones

The Trade-Off Decision

Denormalization makes sense when reads vastly outnumber writes (100:1 or higher ratios). For write-heavy data that changes frequently, the amplification cost may outweigh the read benefits.

Key Takeaway: Denormalization multiplies every write operation across all copies of the data, creating write amplification that increases database load, latency, and operational complexity—making it critical to denormalize only when read patterns justify the write cost.