The words you need, in plain English

Each of these gets a full page later. This list is here so no word on those pages is new to you.

Latency — how long one request takes. Throughput — how many requests you handle per second. They are different things, and making one better often makes the other worse.

Client and server — the client asks (a phone, a browser), the server answers. The request is the question and the response is the answer.

Load balancer — one address in front of several servers. It spreads requests between them and skips any server that stops answering.

Cache — a copy of an answer, kept somewhere faster and closer. Most requests never reach the real source. The price is that the copy can be out of date.

Database — where the truth is kept, on disk, so it survives a restart. Replica — a copy of the database that keeps up with it. You use replicas to survive failure and to serve reads.

Shard, or partition — splitting the data so different machines own different pieces. It is the only way to take more writes than one machine can handle.

Queue — a list of jobs, written down now and done shortly after. It stops the fast part of your system waiting on the slow part.

Stateless — a server that keeps nothing important in its own memory. Any server can handle any request, so losing one costs you only the requests it was handling.

Consistency — whether everyone sees the same data at the same moment. Eventual consistency means they will agree soon, but not right now.

Availability — whether the system answers at all. A system can be up and wrong, or correct and down. Choosing which you prefer, feature by feature, is real design work.

Idempotent — doing it twice has the same result as doing it once. This matters a lot, because retries over a network regularly make things happen twice.