Synchronization systems
for real-time multiplayer.
Custom synchronization systems developed at ITPex to make real-time multiplayer responsive under latency: from tile-based prediction and rollback to deterministic physics for continuous movement.
Thyilea Synchronization Engine
A synchronization engine for multiplayer games with grid- or cell-based movement.
The problem: network latency vs. immediate movement
In a real-time multiplayer game, waiting for the server to confirm every movement makes controls feel delayed. But allowing the client to move without considering authoritative state can produce conflicting game states.
Thyilea addresses this by separating the authoritative game state from the visual representation of that state. The client can therefore respond immediately while the authoritative result is confirmed asynchronously.
Architectural decision: logical state vs. visual state
The synchronization layer does not need to treat the rendered position of a character as its authoritative gameplay state.
A character can already be visually moving toward another cell while the synchronization system still considers the previous cell to be its confirmed logical position. This gives the client freedom to render movement smoothly without changing the state used for authoritative gameplay calculations.
This separation allows immediate visual feedback while keeping gameplay calculations based on discrete, synchronized positions.
Mechanism: prediction and rollback
Prediction becomes important when the local player has already started moving before receiving information about another player's confirmed action.
The local client begins the movement immediately using the information currently available. If a later confirmation reveals that this prediction was invalid, Thyilea reconciles the visual representation with the authoritative state.
The synchronization layer can correct the affected transition while the rest of the gameplay state remains intact. Interpolation keeps remote movement smooth, while the locally predicted character may briefly appear to move back toward its confirmed state.
Edge case: the ghost explosion
Separating visual and logical state also introduces a trade-off under extreme network latency.
A character can visually appear to have reached a safe position while its authoritative logical position is still behind.
If an explosion is evaluated against that logical state before the confirmation catches up, the player can be affected even though the character appears visually outside the explosion.
This is not the normal operating condition of the engine. It illustrates the trade-off introduced by allowing the visual representation to advance before the authoritative state has caught up.
Result: responsive grid-based multiplayer
Thyilea provides a synchronization model in which authoritative gameplay state remains discrete and predictable while visual movement can remain responsive and continuous.
The same separation between state, prediction, confirmation and presentation became the foundation for extending the approach beyond grid-based movement.
Dynagrid Predictive Physics
A lightweight custom physics model designed for deterministic top-down multiplayer gameplay.
The problem: continuous movement
The synchronization principles used by Thyilea worked naturally for games where movement could be represented as transitions between grid cells.
A top-down tank game introduced a different requirement: movement was continuous, and collisions could occur between arbitrary positions rather than discrete cells.
Architectural decision: purpose-built physics
Instead of introducing a general-purpose rigid-body physics system such as Box2D, a custom model was developed around the smaller set of physical behaviors actually required by the game.
The goal was not to reproduce a complete physics engine, but to make movement and collision results deterministic, computationally inexpensive and predictable across clients.
This reduced the amount of state that needed to be reproduced while keeping the physical behavior predictable enough for distributed gameplay.
Mechanism: predicting movement and collisions
Dynagrid predicts future positions from the current state and player input. Instead of simulating a complete rigid-body system, trajectories are represented through linearized segments.
This allows the synchronization layer to reason about future positions and potential collisions before the corresponding state becomes authoritative.
Properties of the model
- Deterministic simulation. Given the same initial state and sequence of inputs, different clients can reproduce the same physical result.
- Piecewise-linear trajectories. Future movement is represented through linearized trajectory segments, making position prediction computationally inexpensive and predictable.
- Distributed physics. Clients can calculate physical movement locally instead of requiring the server to continuously simulate every object's complete trajectory.
- Purpose-built scope. The engine implements only the physics required by the game, avoiding the complexity and overhead of a general-purpose rigid-body system.
Result: distributed prediction
Dynagrid extended the same synchronization philosophy used by Thyilea from discrete grid movement to continuous trajectories.
Clients could calculate future positions locally while maintaining a shared, deterministic interpretation of the authoritative game state.
The networking architecture could support both cell-based movement and continuous physics-driven gameplay without replacing the underlying approach to prediction, confirmation and reconciliation.