PostgreSQL Hashes Lookups: Dynahash and Simplehash
PostgreSQL has two hash tables: dynahash grows one bucket at a time and never rehashes, while simplehash is a templated open-addressing table that resizes without rehashing a …
Performance
PostgreSQL Uses SIMD to Scan Strings Faster: One Vector Layer, Six Consumers
PostgreSQL’s port/simd.h is a portable SSE2/NEON vector layer, and COPY FROM, hex coding, JSON escaping, ASCII validation, linear search and radix-tree lookup all use it to skip …
PostgreSQL Locks Tables Fast: The Fast Path and Sixteen Lock Partitions
Most PostgreSQL locks never touch the heavyweight lock table: weak relation locks live in a per-backend fast path, and the real lock table is split into sixteen …
PostgreSQL Defers and Speeds Up Tuple Deformation: Cached Offsets in TupleDesc
TupleDesc now caches the byte offset of every fixed-width attribute, so fetching column 40 of a wide table can cost one addition instead of a full re-walk.
PostgreSQL Sorts Rows Fast: Abbreviated Keys, Radix Sort, and Top-N Heaps
PostgreSQL’s sorter attacks three problems at once: abbreviated keys make string comparisons into integer compares, a radix sort handles integer keys, and a bounded heap makes …