VLDB Paper Maps Out How Databases Should Actually Write to SSDs

A research paper presented at VLDB digs into the gap between how database systems issue writes and how modern SSDs actually handle them. The authors argue that conventional write patterns inherited from the spinning-disk era leave significant performance and endurance on the table once you account for flash translation layers, garbage collection, and internal parallelism.

The work characterizes how block size, alignment, queue depth, and write hotness interact with SSD firmware behavior, and proposes guidelines for issuing writes that minimize write amplification and tail latency. Benchmarks across a range of commercial drives illustrate where naive append or random-write patterns trigger expensive background work inside the device.

The practical takeaway for storage engine and database authors is that tuning the write path to match SSD internals — rather than treating the device as a black box — can deliver meaningful gains in throughput, latency consistency, and drive lifespan without hardware changes.