about

Measurement-driven systems performance. For people who read the disassembly.

what this site is

lowlat.ms is a systems-performance blog in the Ciechanowski / Giesen / Luu / Lemire lineage: storage, memory, kernel, compute — and the FPGAs that beat them. Every post ships a reproducible methodology block, a real benchmark on real hardware, and citations you can walk yourself.

It refuses a few things on purpose. No hero CTAs. No cookie banners beyond legal minimum. No cover images. No author avatars. No "subscribe to my newsletter" popups. No social share buttons under posts. No JavaScript on the baseline reading path — the Stack Level rail, the Companion glossary tooltips, and the search are CSS and pre-rendered HTML.

how to read it

  • Every post carries a Stack Level rail — ten layers from application down to network. It tells you at a glance what the post actually exercises.
  • Technical terms link into the glossary. Hover a dashed term for the one-sentence definition; click through for the long form and backlinks.
  • The posts index is the firehose. RSS, Atom, and JSON Feed all work.

methodology bar

Every benchmark post lists the CPU, microcode, kernel, compiler, flags, governor, huge pages state, NUMA layout, and a link to a repro repo. If a post cannot name the hardware it ran on, it is not a lowlat.ms post.

contact

Corrections, disagreements, or a benchmark you want reproduced — the repo is on GitHub. Email is hello at lowlat dot ms. I read everything, I answer most things.