Traditional database indexes, like B+ Trees, are brilliant for read-heavy workloads. They keep data perfectly sorted on disk, making lookups incredibly fast.

But this meticulous organization comes at a price.

Every write operation can trigger a cascade of slow, random disk I/O operations as the database must update multiple pages and rebalance the tree structure in place. These small, scattered writes quickly become a bottleneck especially on spinning disks or SSDs under heavy load.

In today’s world of big data, IoT, and real-time analytics, workloads have shifted. Systems are ingesting millions of new records per second. The challenge is no longer just reading efficiently, it’s writing at scale.

We need a data structure that doesn’t just tolerate high write volumes but is built for them.

This is the problem the Log-Structured Merge Tree was designed to solve. It's based on a simple yet powerful principle: stop modifying data in place and start treating writes as an append-only log.

In this chapter, we’ll dive deep into how LSM Trees work, why they’re so effective for write-heavy systems, and how they’ve become the foundation of modern databases like RocksDB, Cassandra, and Bigtable.

1. The Problem with B+ Trees