Benchmarking NVMe Sequential vs Random Queue Depth
2026-09-28 · Storage
At queue depth 1, single-threaded sync writes are bounded by device latency rather than interface bandwidth. Switching batch importers to asynchronous io_uring with modest depth (16–32) improved throughput nearly fourfold without raising tail latency on concurrent reads.
Minimal SQLite Write-Ahead Tuning
2026-08-19 · Databases
For single-node services with bursty writes, enabling PRAGMA journal_mode=WAL together with synchronous=NORMAL and a 5-second busy timeout prevents lock contention while keeping crash recovery straightforward.