Design Online Game Leaderboard

A game leaderboard is a ranked list that shows players their standing compared to others. Most competitive online games provide at least two perspectives: a global ranking of the top performers and a friends-only ranking so players can compare scores with their social circle. The displayed list is short, but the underlying system must sustain rankings efficiently at massive scale while avoiding expensive per-request computation for the friend view.

Requirements

A leaderboard system seems straightforward because it produces only a short sorted list. The real difficulty lies in maintaining rankings cheaply across 50 million players while also serving a friend-scoped view that cannot simply reuse the global ranking query.

Functional Requirements

• A player can submit a new high score or update an existing one after a game session ends.
• Display the global top‑10 leaderboard, which contains the players with the highest scores.
• Display a friend leaderboard for a given player, showing the top‑10 highest-scoring friends.
• Show every player their own global rank and score.

Out of scope (below the line)

• Designing a real‑time multiplayer game server.
• Aggregating leaderboards across multiple games or seasons.
• Anti‑cheat mechanisms or score validation beyond server‑side authority.
• Push notifications when ranks change.
• Time‑windowed leaderboards (daily, weekly, monthly).
• Friend management operations (adding or removing friends).

Non‑functional Requirements

• Leaderboard reads, covering both global and friend views, should complete within 100 ms at the 95th percentile.
• Support 50M registered players, 5M daily active users, and peak score‑update rates around 10K updates per second.
• The global top‑10 should reflect score updates within a few seconds. The friend leaderboard may tolerate slightly higher staleness.
• Reads must remain available during partial failures, favoring slightly stale data over hard errors.
• A player’s own score update must be visible to them immediately (read‑your‑write consistency). Their global ranking position may be eventually consistent.

Back‑of‑the‑envelope Estimation

The numbers that truly drive the design are the 50M entries in the ranking index (far too large for naive SQL rank queries) and the bounded size of a friend list (small enough that reading and sorting friends on the fly is cheap).