Algorithm Visualizations

Algorithm Visualizations

The examples below show the personality of each routing mode. They are not benchmarks; they are a quick way to see whether an algorithm behaves the way you expect.

Legend

Each bar represents the number of players on a server. The goal is to see how evenly (or intentionally unevenly) each algorithm spreads the load.

█ = ~5 players
▌ = ~2-3 players

---

Low Load — 5 Players, 3 Servers

least_players

lobby-1: ██   (2)
lobby-2: ██   (2)
lobby-3: █    (1)

power_of_two

lobby-1: ██   (2)
lobby-2: ██   (2)
lobby-3: █    (1)

round_robin

lobby-1: ██   (2)
lobby-2: ██   (2)
lobby-3: █    (1)

random

lobby-1: ███  (3)
lobby-2: █    (1)
lobby-3: █    (1)

• Higher variance at low counts — expected behavior.

weighted_round_robin (weights: 3, 2, 1)

lobby-1: ███  (3)
lobby-2: █    (1)
lobby-3: █    (1)

least_connections

lobby-1: ██   (2)
lobby-2: ██   (2)
lobby-3: █    (1)

consistent_hash

lobby-1: ██   (2)
lobby-2: ██   (2)
lobby-3: █    (1)

• Depends on UUID hash; shown here as a typical distribution.

---

Medium Load — 30 Players, 5 Servers

least_players

lobby-1: ██████   (6)
lobby-2: ██████   (6)
lobby-3: ██████   (6)
lobby-4: ██████   (6)
lobby-5: ██████   (6)

• Near-perfect even distribution.

power_of_two

lobby-1: ██████   (6)
lobby-2: ██████   (6)
lobby-3: ██████   (6)
lobby-4: ██████   (6)
lobby-5: ██████   (6)

• Very close to least_players — at this scale, practically identical.

round_robin

lobby-1: ██████   (6)
lobby-2: ██████   (6)
lobby-3: ██████   (6)
lobby-4: ██████   (6)
lobby-5: ██████   (6)

• Strictly even by design.

random

lobby-1: ████████ (8)
lobby-2: █████▌   (5)
lobby-3: ██████   (6)
lobby-4: ████▌    (4)
lobby-5: █████▌   (7)

• Some variance — evens out further as the player count grows.

weighted_round_robin (weights: 5, 4, 3, 2, 1)

lobby-1: ██████████   (10)
lobby-2: ████████     (8)
lobby-3: ██████       (6)
lobby-4: ████         (4)
lobby-5: ██           (2)

• Proportional to weight.

least_connections

lobby-1: ██████   (6)
lobby-2: ██████   (6)
lobby-3: ██████   (6)
lobby-4: ██████   (6)
lobby-5: ██████   (6)

• EMA smoothing produces an even distribution under steady load.

consistent_hash

lobby-1: ██████   (7)
lobby-2: ██████   (6)
lobby-3: █████    (5)
lobby-4: ██████   (6)
lobby-5: ██████   (6)

• Minor variance from hash ring distribution.

---

High Load — 100 Players, 10 Servers

least_players

srv-01: ██████████ (10)
srv-02: ██████████ (10)
srv-03: ██████████ (10)
srv-04: ██████████ (10)
srv-05: ██████████ (10)
srv-06: ██████████ (10)
srv-07: ██████████ (10)
srv-08: ██████████ (10)
srv-09: ██████████ (10)
srv-10: ██████████ (10)

power_of_two

srv-01: ██████████ (10)
srv-02: ██████████ (10)
srv-03: █████████▌ (11)
srv-04: █████████  (9)
srv-05: ██████████ (10)
srv-06: ██████████ (10)
srv-07: █████████▌ (11)
srv-08: █████████  (9)
srv-09: ██████████ (10)
srv-10: ██████████ (10)

• ±1 deviation — performs well at scale.

round_robin

srv-01: ██████████ (10)
srv-02: ██████████ (10)
srv-03: ██████████ (10)
srv-04: ██████████ (10)
srv-05: ██████████ (10)
srv-06: ██████████ (10)
srv-07: ██████████ (10)
srv-08: ██████████ (10)
srv-09: ██████████ (10)
srv-10: ██████████ (10)

random

srv-01: ███████████ (12)
srv-02: █████████▌  (9)
srv-03: ██████████  (10)
srv-04: ████████▌   (8)
srv-05: ███████████ (11)
srv-06: ██████████  (10)
srv-07: █████████▌  (9)
srv-08: ██████████  (10)
srv-09: ███████████ (11)
srv-10: ████████▌   (10)

• Variance shrinks with scale — law of large numbers.

weighted_round_robin (weights: 5, 5, 3, 3, 3, 2, 2, 2, 1, 1)

srv-01: ██████████████████  (19)
srv-02: ██████████████████  (19)
srv-03: ███████████         (11)
srv-04: ███████████         (11)
srv-05: ███████████         (11)
srv-06: ████████            (8)
srv-07: ████████            (8)
srv-08: ████████            (8)
srv-09: ████                (4)
srv-10: ████                (4)

least_connections

srv-01: ██████████ (10)
srv-02: ██████████ (10)
srv-03: ██████████ (10)
srv-04: ██████████ (10)
srv-05: ██████████ (10)
srv-06: ██████████ (10)
srv-07: ██████████ (10)
srv-08: ██████████ (10)
srv-09: ██████████ (10)
srv-10: ██████████ (10)

consistent_hash

srv-01: █████████▌ (11)
srv-02: ██████████ (10)
srv-03: █████████  (9)
srv-04: ██████████ (10)
srv-05: █████████▌ (11)
srv-06: █████████  (9)
srv-07: ██████████ (10)
srv-08: ██████████ (10)
srv-09: █████████▌ (10)
srv-10: █████████  (10)

• Hash ring produces ±1 deviation across 10 servers.

---

Latency Mode — Proxy-to-Backend Measurement

latency does not produce an even distribution chart. It ranks the healthy candidates by the ping measured from the Velocity proxy:

lobby-east:  25 ms  ← selected
lobby-west:  70 ms
lobby-eu:   110 ms

Every player routed by that proxy sees the same current ranking. This is useful when backend network delay differs, but it is not player-specific geographic routing. If the 25 ms server becomes full, drained, unhealthy, circuit-open, or lifecycle-disallowed, the next eligible candidate can be selected.

---

Key Takeaways

Scenario	Best Algorithm
Want perfect balance, don't care about cost	least_players
Want near-perfect balance, low cost	power_of_two
Servers have different capacities	weighted_round_robin
Need sticky sessions	consistent_hash
Bursty traffic patterns	least_connections
Prefer the lowest proxy-to-backend ping	latency
Just testing / strict fairness	round_robin
Very large server pool (50+)	random

See Routing Algorithms for detailed explanations of each mode.