One wala, many kettles (concurrency vs parallelism)

The stall on the corner, its queue and its wala at the stove. This lesson stays inside that one stall and asks what the wala does while the chai brews. The lab at the end of the story lets you run the same stall yourself.

Before you start. You need the stall from The chai stall that got famous (load balancing): one wala at one stove, a queue that grew when the stall got famous, and more stalls added so the crowd could be shared. You do not need to know anything about how computers work; every word this lesson uses is shown doing something before it is named.

The idea in 30 seconds

A web server spends most of each request waiting on something else, usually the database, not working.
A worker that waits for one request to finish before starting the next is blocking, and wastes all that waiting.
Concurrency means keeping many requests in progress and switching to whichever is ready; it finishes a batch sooner but makes no single request faster.
An event loop is one worker going round a single rule: do whatever is ready, then look again. One long piece of real work stalls every request behind it.
Parallelism means more workers running at the same moment, and it only helps once one worker is busy all the time.
Keeping many requests in progress needs a cap on how many, because the thing being waited on slows under load, and per-server caps add up across every server.

Read the story · Skip to the numbers