Skip to content

SimulationConfigGuide.md

Latest commit

 

History

History
219 lines (170 loc) · 11.9 KB

SimulationConfigGuide.md

File metadata and controls

219 lines (170 loc) · 11.9 KB

SimulationConfig Recommended-Value Guide (per Genre / Platform)

This document is a starting-point guide for how to configure the key parameters of ISimulationConfig / SimulationConfig by game genre and platform.

Recommended values are only a starting point. Tune them against measured RTT/jitter, content-driven input frequency, and concurrent entity counts. Default definitions live in SimulationConfig.cs; field semantics are documented in Specification.md §2.2.

1. Key Parameters and Their Impact

Parameter Smaller Larger Determinism Impact
TickIntervalMs Faster response · network load ↑ · simulation cost ↑ Slower response · load ↓ ✅ must be identical across all peers
InputDelayTicks More direct feel · less network slack Better jitter absorption · higher input latency ✅ must be identical
MaxRollbackTicks Less memory · narrower rollback window More memory · wider recovery window on hiccups ✅
SyncCheckInterval Faster desync detection · more traffic Less traffic · slower detection —
UsePrediction Pause on missing input (Paused) Predict, then rollback —
SDInputLeadTicks Less server-arrival slack · risk of unapplied input Higher perceived input latency · greater stability — (SD only)
InterpolationDelayTicks Fresher remote entities · more jitter exposure Smoother interpolation · more delay — (View)

Key formulas (SD mode):

Effective input latency (perceived) ≈ (InputDelayTicks + SDInputLeadTicks) × TickIntervalMs
Server input arrival deadline       ≈ HardToleranceMs (auto: above + RTT/2 + 20 ms jitter)

Key formulas (P2P mode):

Effective input latency (perceived) ≈ InputDelayTicks × TickIntervalMs
Remote input arrival slack          ≈ InputDelayTicks × TickIntervalMs (RTT/2 must be < this for hiccup-free play)

2. Mode Selection Guide

Game Type Recommended Mode Rationale
1v1 fighting, small co-op (≤4) P2P Direct RTT, no host overhead
Competitive PvP / matchmaking / ranked ServerDriven Cheat prevention, matchmaking-friendly
Mobile matched PvP ServerDriven Mobile NAT traversal / stability
Synchronous turn-based (≥4) ServerDriven Input authority, end-of-match verification
Asynchronous (single-player / replay) P2P (solo) Network not required

Klotho's two modes have different message flows and session-entry paths, so the mode is usually fixed once the game genre is decided.

3. Per-Genre Recommended Values

Notation:

  • PC = desktop / console (assumes stable RTT 30–80 ms)
  • Mobile = LTE/5G (assumes RTT 60–150 ms · jitter 20–60 ms)
  • Bold values are recommended starting points; parentheses give the acceptable range
  • All ms values are integers

3.1 Fighting — 1v1 / 2v2

A genre where a single input frame translates directly to a visible result. Rollback is the top priority with minimal input latency.

Parameter PC Recommended Mobile Recommended Note
Mode P2P ServerDriven PC = direct RTT; mobile prioritizes stability
TickIntervalMs 17 (16–25) 25 (25–33) PC=60 Hz / Mobile=40 Hz
InputDelayTicks 2 (1–3) 3 (2–4) PC ~34 ms / Mobile ~75 ms
MaxRollbackTicks 8 (6–12) 10 (8–14) Standard for fighting rollback
SyncCheckInterval 30 30 Once per second (Mobile ~1.2 s)
UsePrediction true true Required
SDInputLeadTicks — 2 (2–4) Keep small under SD
EnableErrorCorrection false true Smoother interpolation on mobile
InterpolationDelayTicks 1 2 Short — responsiveness first
MaxEntities 32 32 Sized to characters + projectiles

Tuning notes:

  • If perceived input latency is too high on mobile, drop InputDelayTicks to 2 but raise MaxRollback to 12 to compensate.
  • For 1v1, MaxRollback can be reduced to 8 to save memory.

3.2 Action / Brawler / FPS · TPS (PvP)

Many entities · responsiveness matters · deterministic simulation (the Brawler sample falls in this category).

Parameter PC Recommended Mobile Recommended Note
Mode ServerDriven ServerDriven Cheat prevention
TickIntervalMs 25 (25–33) 40 (33–50) PC=40 Hz / Mobile=25 Hz
InputDelayTicks 0 (0–2) 0 (0–2) Recommended 0 under SD (Lead absorbs it)
MaxRollbackTicks 50 (30–60) 50 (30–60) Rollback depth ≈ 1.25–2 s
SyncCheckInterval 30 30
UsePrediction false false SD is server-authoritative — prediction not needed
SDInputLeadTicks 4 (3–6) 6 (4–10) Higher on mobile to absorb RTT
HardToleranceMs 0 (auto) 0 (auto) Auto reflects RTT
InputResendIntervalMs 150 (100–200) 150 (100–250) Unacked-input resend
MaxUnackedInputs 30 30
EnableErrorCorrection true true Remote-entity correction
InterpolationDelayTicks 2 3 Higher on mobile to absorb jitter
MaxEntities 256 128 Proportional to content scale

The Brawler sample (Tools/BrawlerDedicatedServer/simulationconfig.json) is a real-world PC example for this category.

Tuning notes:

  • For builds with many characters/projectiles, set MaxEntities to the measured peak + 25% headroom.
  • In mobile handover environments (LTE↔5G transitions), raise SDInputLeadTicks to 8–10.
  • When running P2P, raise InputDelayTicks to 3–4 and ignore SDInputLeadTicks.

3.3 MOBA / Co-op Action RPG (PvE)

Parameter PC Recommended Mobile Recommended Note
Mode ServerDriven ServerDriven
TickIntervalMs 33 (25–50) 50 (40–66) PC=30 Hz / Mobile=20 Hz
InputDelayTicks 0 0 SD absorbs via Lead
MaxRollbackTicks 30 (20–50) 30 (20–50)
SyncCheckInterval 30 30
UsePrediction false false
SDInputLeadTicks 4 (3–6) 6 (4–10)
EnableErrorCorrection true true
InterpolationDelayTicks 2 3
MaxEntities 512 256 Includes minions/mobs

3.4 RTS (Real-Time Strategy)

Many units · low input frequency · rollback impractical (simulation cost too high). Fixed lockstep + prediction disabled pattern.

Parameter PC Recommended Mobile Recommended Note
Mode P2P or ServerDriven ServerDriven Cheat prevention / matchmaking
TickIntervalMs 100 (66–200) 150 (100–250) 10 Hz–5 Hz
InputDelayTicks 6 (4–10) 8 (6–12) Tolerates 600–1200 ms input latency
MaxRollbackTicks 2 (1–4) 2 (1–4) Effectively no rollback
SyncCheckInterval 10 10 Fast desync detection
UsePrediction false false Required
SDInputLeadTicks 4 6 Only under SD
EnableErrorCorrection false false Correction unnecessary
MaxEntities 1024+ 512+ Sized to unit count

Tuning notes:

  • A small SyncCheckInterval is fine despite the traffic cost (the absolute frequency is low because ticks themselves are slow).
  • When raising MaxEntities, also measure ECS component memory and hash cost.

3.5 Tactics / Strategy — Non-Real-Time

Each turn is discrete; simulation advances after input arrives. Slow tick + no rollback.

Parameter PC Recommended Mobile Recommended Note
Mode ServerDriven ServerDriven Matchmaking / verification
TickIntervalMs 100 (50–250) 200 (100–500) 5 Hz–2 Hz
InputDelayTicks 2 (0–6) 2 (0–6) Turn-based games tolerate large latency
MaxRollbackTicks 2 (1–4) 2 (1–4) No rollback
SyncCheckInterval 5 5
UsePrediction false false
SDInputLeadTicks 2 2
EnableErrorCorrection false false
MaxEntities 256 128 Sized to board

4. Platform / Environment Adjustment Guide

4.1 Mobile (LTE / 5G / Wi-Fi)

  • RTT variability: assume +50–100 ms average RTT and +20–40 ms jitter standard deviation versus PC.
  • SDInputLeadTicks: add +2–4 to the PC recommendation.
  • InterpolationDelayTicks: add +1 (jitter absorption).
  • TickIntervalMs: bump up one tier vs. PC (e.g., PC 25 ms → Mobile 40 ms).
  • Battery / heat: lowering tick rate reduces both simulation cost and send/receive frequency, so it has a large impact.

4.2 Console (wired / stable)

  • PC recommendations work as-is. TickIntervalMs can be reduced to 16–17 (60 Hz) for fighting/action games.

4.3 Global Matchmaking (cross-region RTT > 150 ms)

  • Recommend SDInputLeadTicks ≥ 8.
  • Keep HardToleranceMs at 0 (auto) — it accounts for RTT and is safer.
  • Add +1–2 to the recommended InterpolationDelayTicks.

5. Diagnosis / Tuning Workflow

  1. Measure before locking values: enable basic diagnostics with TickDriftWarnMultiplier=2 and EventDispatchWarnMs=5.
  2. Watch the logs:
    • Tick gap warnings → revisit TickIntervalMs suitability or simulation cost.
    • Accumulator clamped → frame drops or TickIntervalMs set too low.
    • SD MaxUnackedInputs exceeded → adjust InputResendIntervalMs or raise SDInputLeadTicks.
  3. On desync: temporarily lower SyncCheckInterval (e.g., 10 → 5) for faster detection.
  4. Interpolation jitter: raise InterpolationDelayTicks one step at a time and enable EnableErrorCorrection.
  5. On OnMatchAborted(ChainStallTimeout) false positives: verify SessionConfig.ReconnectTimeoutMs against the recovery floor. Effective threshold = max(ReconnectTimeoutMs / TickIntervalMs + 100, SessionConfig.MinStallAbortTicks). If ReconnectTimeoutMs < 30s, MinStallAbortTicks (default 600 = 30s @ 50ms) acts as the floor — raise both for high-latency/long-recovery scenarios.

Field placement — MinStallAbortTicks, LateJoinDelaySafety, RttSanityMaxMs live in SessionConfig alongside the other LateJoin/Reconnect/ChainStall service-side knobs. The engine-internal knobs — CatchupMaxTicksPerFrame, ResyncMaxRetries, DesyncThresholdForResync, CorrectiveResetCooldownMs, Reactive Dynamic InputDelay 6 (ReactiveWindowTicks, ReactiveEscalateThreshold, ReactiveStep, ReactiveMax, ServerPushGraceTicks, ReactiveEscalateCooldownTicks), and Rollback Burst 2 (RollbackBurstCount, RollbackWindowTicks) — live in SimulationConfig.

6. Summary — Decision Tree

1) Is the game matchmaking/ranked-based?
   YES → ServerDriven
   NO  → P2P (small friends co-op)

2) Is visible 60 Hz input response important? (fighting · precision shooter)
   YES → TickIntervalMs ≤ 25, InputDelayTicks ≤ 3, MaxRollback 8–12, UsePrediction=true
   NO  → next step

3) Character-scale action, or many-unit strategy?
   Character     → TickIntervalMs 25–40, MaxRollback 30–50
   Many units    → TickIntervalMs 100–200, MaxRollback 1–4, UsePrediction=false

4) Mobile build?
   YES → TickIntervalMs +1 tier, SDInputLeadTicks +2–4,
         InterpolationDelayTicks +1, EnableErrorCorrection=true

7. References