SSTables and LSM-Trees: The Secret Sauce Behind Modern Database Performance

Description: An explanation of how databases like Cassandra and RocksDB use Log-Structured Merge Trees and Sorted String Tables to handle millions of writes per second.

Date: August 24, 2025

Traditional databases often struggle with a classic trade-off: optimize for writes, and reads suffer. LSM Trees take a different approach — they prioritize write performance and accept that reads will be slower.

What Are SSTables?

Sorted String Tables (SSTables) are key-value files where the keys are stored in sorted order. SSTables are immutable and append-only — once written to disk, they never change. This makes writes incredibly fast because they are pure sequential I/O.

The LSM Tree Workflow

1. The Memtable (In-Memory Writes): All write operations initially go to an in-memory balanced tree, which is blazing fast.
2. Flushing to Disk: When the memtable reaches a threshold size, it is flushed to disk as a new, sorted SSTable.
3. Reading Data: Reads check the memtable first, then the most recent SSTables, then older ones. This can be slower than B-tree approaches.
4. Background Compaction: Periodically, smaller SSTables are merged into fewer, larger ones to optimize storage and read performance.

Smart Optimizations

To avoid searching every SSTable for a key that doesn’t exist, a Bloom filter is used. It can tell you with 100% certainty if a key does not exist in a data file, preventing the worst-case search scenario.

Read the full, original article on LinkedIn.