Each of these parameters controls a different resource, and the classic mistake is treating them as one big "make it faster" dial. Know precisely what each governs and which ones multiply.
shared_buffers
Postgres's own page cache — the shared-memory region where it keeps recently touched table and index pages. Reads check here first; misses fall through to the OS page cache and then disk. Rule of thumb: ~25% of RAM. Going much higher rarely helps because Postgres also benefits from the OS cache (double buffering), and very large values can hurt checkpoint behavior.
work_mem
The memory budget for a single sort or hash operation. The trap: it is allocated per node, per connection — one complex query with three sorts and two hash joins can use 5 × work_mem, and 100 connections doing that simultaneously multiplies again. Set conservatively (e.g. 16–64 MB) and raise it locally with SET work_mem for known-heavy reporting queries. Exceeding it spills sorts/hashes to temp files on disk.
maintenance_work_mem
Like work_mem but for maintenance operations: VACUUM, CREATE INDEX, ALTER TABLE ... ADD FOREIGN KEY. These run rarely and one at a time per session, so it's safe to set it large (e.g. 512 MB–2 GB) — bigger values make index builds and vacuums substantially faster.
effective_cache_size
A planner hint, not an allocation — it tells the planner how much memory (Postgres buffers plus OS cache) is likely available for caching. It changes nothing about actual memory use; it only nudges the planner toward index scans when it believes pages are probably cached. Set to ~50–75% of RAM.
max_connections
The hard cap on concurrent connections. Each connection is a separate backend process with its own memory overhead, so raising this is expensive and is not the way to scale. Use a pooler (PgBouncer) instead and keep this modest.
random_page_cost
The estimated cost of a random page read relative to seq_page_cost = 1.0. Default 4.0 assumes spinning disks; on SSD/cloud storage lower it to ~1.1 so the planner correctly sees index scans as cheap.