Skip to main content

Writing

Engineering notes on HTTP infrastructure, proxies and web performance: what we measured, how, and what it cost.

  1. Envoy CPU per request: what held up, what we dropped

    Where Envoy spends CPU per request, what we cut and what we withdrew: route matching, PGO and ThinLTO, a settings price list and the method behind it.

  2. Envoy with PGO and ThinLTO: 20% to 40% less CPU per request

    An Envoy built with PGO and ThinLTO used 20% to 40% less CPU per request than a standard build of the same commit. Numbers, method, costs and recipe.

  3. What Envoy settings and releases cost in CPU: a measured price list

    CPU per request for nine Envoy settings, a second hop, three releases, surplus worker threads and one experimental change, with conditions and 95% intervals.