Every multiplayer game you have ever blamed for lag was making one architectural decision visible: who holds the truth? When a WoW raid boss casts a spell, when a StarCraft II marine fires a shot, and when an extraction shooter decides whether your bullet landed before your death — those are all answers to the same question. This is the plain-language explainer of client-server versus peer-to-peer synchronization, why MMOs chose the architecture they did, and how 2026’s live games are bending the old rules with server meshingserver meshing. A technique that stitches many game servers together so players see one seamless world, even across server boundaries..

What it covers: The client-server model (server authority, tick rates, interest management), peer-to-peer and lockstep alternatives, why genres cluster on specific architectures, and how server meshing changes the ceiling. Why it matters: The netcode model explains genre behavior you already feel — MMOMMO. Massively Multiplayer Online — a game that hosts very large numbers of players together in one shared, persistent world. rubber-banding, RTSRTS. Real-Time Strategy — a strategy game where everyone plays at the same time, continuously, instead of taking turns. desyncs, extraction-shooter peeker’s advantage — and it is the real cost behind “massive” promises. Who should pick this: Players who want to understand why 100-player zones work but 10,000-player battles never have, and anyone following Star Citizen’s server meshing or MMO scalability claims.

The core question: who holds the truth?

Client-server puts one authoritative machine in charge. Every player’s client sends inputs (“I moved here, I cast this”) to the server; the server decides what actually happened and broadcasts the resulting state back. Your client is a renderer with an opinion — the server is the only one allowed to be right. This is the model behind virtually every MMO: WoW’s zones, FFXIV’s data centers, Once Human’s 256 km² maps all run server-authoritative simulation.

Peer-to-peer removes the referee. Every client talks to every other client (or a designated “host” player machine acts as a lightweight server) and each machine decides what happened on its own screen. Nothing is authoritative — the game is a negotiated agreement between peers. Cheating becomes trivially possible because the machine claiming “I hit you” is the same machine deciding whether it hit.

Client-serverPeer-to-peer
Truth holderDedicated serverEach peer (or host player)
Cheat resistanceHigh — server validatesLow — clients self-report
Latency feelConsistent for allHost advantage (host ≈ 0 ping)
Scale ceilingHardware-bound but highDegrades fast with peer count
Operating costYou pay for servers foreverFree — players host
Typical useMMOs, shooters with ranked playFighting games, older RTS, co-op

How servers scale to thousands: ticks, zones, and interest management

A server processing 5,000 players does not simulate 5,000 players against each other. Three techniques make the “massively” in MMO honest:

  1. The tick loop. The server simulates the world in fixed steps — a tick rate of 10–30 Hz is typical for MMOs ( shooters run higher; MMOs trade frequency for world size). Everything between ticks is interpolation smoothing on your client.
  2. Zoning/sharding. Players are partitioned onto separate server processes — WoW’s zones are literally separate simulations, and “layering” or “phasing” means two players in the same place may be in different instances of the same place.
  3. Interest management. The server only tells you about what can affect you. The other factionfaction. An in-game group or nation players can align with or fight against, with its own goals and reputation system.’s auction house crowd is not data you receive; your client renders what the server decided you need to know. This is why MMO population feels invisible past render distance — it is mathematically invisible.

The cost of server authority is latency: your input travels to the server, the server processes the tick, the result travels back. That round trip is why MMO movement has historically felt floatier than single-player games — the client shows your movement immediately (client-side prediction) and the server corrects any disagreement, which you experience as rubber-banding when the correction disagrees.

The lockstep alternative (and why RTS games use it)

StarCraft II and classic RTS games run deterministic lockstep: every player’s client receives every other player’s inputs and simulates the identical game from them. No truth-holder exists — instead, all clients must compute bit-identical results from the same inputs, and any divergence is a desync (the game’s own error term). The upside is bandwidth perfection: you only transmit commands, not world stateworld state. The shared record of everything that has happened in the game world — who owns what, what's been built or destroyed., which is why a 2010 game handles huge armies over bad connections. The downside is that a desync breaks everything, and there is no referee to say who was right — the community “maphack” and “drop-hack” problems of RTS history are lockstep’s architectural children. The June 2026 patch 5.0.16 economy rework covered in our SC2 primer changes nothing about this layer — the simulation remains lockstep, which is why ladder integrity still depends on client validation.

Fighting games, the other lockstep holdout, pair determinism with rollback netcode: peers predict inputs and roll the simulation back when predictions were wrong — the technique that turned genre netcode from a joke into a solved problem, and the reason it only works on games built for it.

Why genres cluster where they do

  • MMOs (client-server): anti-cheat is existential — a duped item is an economy collapse; persistence requires one truth; scale requires zoning anyway.
  • Competitive shooters (client-server, with server-side hit validation): ranked integrity requires a referee; peeker’s advantage is the accepted cost of prediction.
  • RTS/fighting (lockstep/P2P): unit counts make state transmission impossible; determinism is the only affordable bandwidth strategy.
  • Extraction shooters (session client-server, increasingly): Tarkov-style raids run small authoritative servers per raid — the 2026 survival wave (ARC Raiders, Icarus prospects) applies the same small-server discipline to survival, which is why their “MMO-adjacent” scale claims are really many small matches, not one big world.

The frontier: server meshing

The old ceiling — one authoritative process per zone — is what Star Citizen has spent years attacking with server meshing: multiple authoritative servers dynamically handing players and objects between each other so that one seamless world spans many machines. The engineering reality is a distributed-consensus problem (what happens at the moment of handoff? whose physics wins when two players straddle two servers?) — covered in detail in our Star Citizen server meshing explainer. The promise is that “number of players per zone” stops being a hardware constant and becomes a budgeting decision; the catch is a decade of engineering to make handoffs invisible. When a 2026-era pitch says “thousands of players in one battle,” it is either meshing, instances with a marketing budget, or a lie — and you can now tell which.

The takeaway

The next time a game feels laggy, ask who holds the truth. Rubber-banding in an MMO is the server correcting your prediction — working as designed. A host player winning every 50/50 duel is P2P host advantage — working as designed, badly. A strategy game ending in a desync is lockstep’s determinism breaking — the architecture doing the only thing it can. Netcode is not an implementation detail; it is the physics of whose game you are actually playing.

Next read

SOURCES