Tag: performance

Leave most autovacuum globals alone, raise the cost limit, and give your one hot table a per-table scale_factor. Tested on postgres:18.6 with real numbers.

A 125B MoE ran an agent job locally for 72 hours and made real progress. A free 120B cloud route finished the same job in 7. Here is where the time went.

One Go shop app as a monolith and as gRPC microservices on the same pinned CPU cores. The monolith won every throughput test, and Postgres capped some loads.

Every chat turn re-sends the whole history. See the token math, what prompt caching saves on the API, and what a long context costs in VRAM when you run local.

Why apps fall over under load: pool size times replicas versus Postgres max_connections, restart herds, retry storms, timeouts, and the configs that fix them.

zstd -3 compresses 4 to 6x faster than gzip -6 with smaller files. At the top end, xz -6 beat zstd -19 on speed. 19 configs, 4 public data sets, measured.

Stop watching osm2pgsql crawl. Tune flex output, slim mode, --cache, and Postgres for OSM imports on hardware that isn't a hyperscaler.

ZFS tuning for flash: ARC sizing, recordsize, autotrim, sync modes, and SLOG advice, plus why defaults written for spinning disks waste an NVMe pool.

PREEMPT_RT hit Linux mainline in 2024. Here's what real-time actually means, when you need it, and when it'll just slow you down.

Linux PSI (Pressure Stall Information) exposes real CPU, memory, and IO stall percentages, here's why it beats load average for monitoring.

Build Docker images that run on amd64 and arm64 from one command. QEMU emulation, native multi-node builders, CI caching, and manifest lists explained.

cgroups v2 unified hierarchy explained: read /sys/fs/cgroup/, limit CPU/IO/memory, use systemd slices and PSI pressure files.