CSE Roadmap

Core tier · about 2 hours · 6 parts

Part 2 of 6 · 2 min read

Describing load

Think of load parameters like the vital signs a doctor checks. Heart rate alone does not describe a patient; the right handful of numbers, together, does.

In plain words: Load is described with a small number of load parameters, chosen to fit the specific system. For a web server that might be requests per second; for a database it might be the ratio of reads to writes; for a chat system it might be the number of simultaneously active users. Sometimes a single load parameter hides the real difficulty. A system that both writes data and fans that data out to many readers has a fan-out load parameter: how many downstream writes or reads one single incoming request triggers. A system with a large, uneven fan-out can struggle even when its raw write rate looks small.

A worked example. The bookstore's new-release notification feature sends one email per follower whenever an author publishes a book. Most authors have a handful of followers, but one bestselling author has 2 million. A single "new book" event becomes 2 million outgoing emails; the real load parameter is not "books published per day" but "followers per author," since that is what determines the actual work.

The key lines. No new SQL; this is a measurement, not a query.

Load parameters observed:
  books_published_per_day:  40
  followers_per_author:     1 to 2,000,000 (highly skewed)
  resulting_email_sends:    up to 2,000,000 per single publish event

Load parameter, fan-out

A load parameter is a specific, measurable number chosen to describe how busy a system is, picked to fit that system's own bottleneck rather than a generic count. Fan-out is the number of downstream operations, writes or reads to other systems or users, that a single incoming request triggers.

Example: "followers per author" is the load parameter that actually predicts the bookstore notification system's real workload, not the smaller and steadier "books published per day."