Skip to content

This is general knowledge or an external-engine reference, not an Apollo architecture or implementation claim. See Apollo's architecture status for target terminology and verified implementation boundaries.

Q50: How do you design a matchmaking system? ​

Core Conclusion ​

A matchmaking system really has to balance three things:

  • Fairness
  • Waiting duration
  • Explainability

There is no matchmaking algorithm that is always optimal; there are only trade-offs more suitable for a specific gameplay and online scale.

If you only stare at "closest strength", the waiting time gets longer; if you only stare at "instant start", the match quality is very poor. A matchmaking system is essentially a dynamic balancer.

1. First clarify what the matching object is ​

Different gameplay matches very different objects:

  • Single player matching single player
  • Team matching team
  • Multi-character class composition matching
  • PvP rank matching
  • PvE team-up matching

If the object model is not defined clearly, the later scoring and team-up logic will be chaotic.

2. Which dimensions a matchmaking system really cares about ​

Common ones include:

  • Strength or rank
  • Waiting time
  • Team size
  • Class composition
  • Server or region
  • Network quality

Not every gameplay needs the full set of dimensions, but the primary and secondary must be determined first.

3. Fairness and waiting duration are necessarily in conflict ​

This is the basic reality of a matchmaking system.

Stricter matching conditions mean:

  • Closer matches
  • Longer waits

Looser matching conditions mean:

  • Faster starts
  • More unstable match quality

So many systems adopt the strategy of "gradually widening the match window".

4. Matchmaking pool design matters more than formulas ​

Many people focus all their attention on the scoring algorithm, but what is actually more important is often matchmaking pool management.

What needs to be clear:

  • How players enter the pool
  • How teams are represented
  • How timeouts are handled
  • How to return to the pool after accept-confirmation failure

If this part is not designed well, even the best scoring function cannot land.

5. Different gameplay suits different matchmaking focuses ​

1. Competitive PvP ​

Emphasizes more:

  • Close strength
  • Fair teams
  • Reasonable rank and hidden score

2. PvE team-up ​

Emphasizes more:

  • Start speed
  • Complete role responsibilities
  • Basic combat power threshold

3. Large battlefields or events ​

Emphasizes more:

  • Quickly gathering headcount
  • Faction balance
  • Cross-server aggregation capability

6. After a successful match, the "accept phase" still needs handling ​

Many systems do not enter directly after matching, but instead have:

  • Accept confirmation
  • Timeout cancellation
  • Fill-in for missing players

This phase is important because it can significantly reduce the waste of "instantly abandoning after being matched".

7. Cross-server matchmaking introduces extra constraints ​

Once cross-server, the system usually also has to consider:

  • Region/server latency
  • Language or region
  • Event time windows
  • Instance server carrying capacity

So cross-server matchmaking is not only enlarging the pool, but also enlarging scheduling complexity.

8. A more robust engineering design ​

Common practice is:

  • First build a clear matchmaking object model
  • Use segmented scoring and gradual window widening to control fairness and waiting
  • Enter the accept confirmation phase after a successful match
  • Finally hand off to the instance or battlefield instance system to carry

This way the boundary between matchmaking and instance creation is also clearer.

9. Common misconceptions ​

1. The more complex the matchmaking algorithm, the better ​

No. What matters more in many cases is that rules are explainable, tunable, and observable.

2. As long as the hidden score is accurate, matchmaking will have no problems ​

No. Team composition, class structure, regional latency, and pool size all affect the experience.

3. A matchmaking system is just a sorting algorithm ​

Far from it. It is also a queue system, a scheduling system, and an experience balancing system.

References ​

  • Various PvP / PvE matchmaking systems, hidden score, and window-widening strategy practice materials

Released under the MIT License