Elevator System

Prompt

"Design the software for the elevators in an office building. Several cars, a few dozen floors, buttons in the hall and inside the cars. Show me the objects and how you decide which car answers a call."

Clarifying questions

• How many cars and floors? Say 4 cars and 30 floors. Small enough that any scheduler is fast; large enough that a bad scheduler is visibly bad.
• What inputs exist? Hall calls (floor + direction) and car calls (a floor pressed inside a specific car). Door open/close and emergency stop are follow-ups.
• Is this real-time control or a simulation? A discrete simulation: the system advances in ticks, each tick a car moves one floor or spends one tick with doors open. The interviewer wants to see the state machine advance, not talk to hardware.
• What is the goal of the scheduler? Minimize average wait time, avoid starvation. No express or zoned cars in the first version.
• Can calls arrive while the system is stepping? Yes. Buttons are pressed from other threads (or callbacks) while the tick loop runs, so the request intake must be thread-safe.
• Capacity limits? Ignore weight. Mention that a full car should skip hall calls.

Requirements

Functional

• Accept hall calls and car calls at any time.
• Assign each hall call to exactly one car; a car serves its stops in a sensible order.
• Advance the whole system by one tick: move cars, open and close doors, clear served stops.
• Expose car state (floor, direction, door) for a display.

Non-functional

• No lost calls and no call served twice.
• No starvation: a call placed at any floor is eventually served even under continuous load.
• Scheduling cost per call must be small: O(cars) or O(cars x floors) is fine at this size.
• Scheduler policy must be replaceable, because "compare two policies" is the standard follow-up.

Numbers

• 4 cars x 30 floors = at most 120 distinct pending stops. One tick per second of simulated time. Even a naive O(cars x floors) evaluation per call is a few hundred operations, which frees you to choose the clearest algorithm rather than the cleverest.