Wondering what desync, tick rate or fill rate actually mean for your game? Here's key multiplayer term, explained by the team who run servers for millions of players.
Multiplayer infrastructure has a vocabulary problem: the same word often means three different things depending on who's using it (*cough* lobbies *cough*).
Browse the definitions covering game server hosting, orchestration, netcode, matchmaking and network architecture.
Infrastructure & Costs
The machines, locations and pricing models related to multiplayer game server.
A bare metal server is a whole physical machine, rented by the month. What it costs a multiplayer game, how it differs from cloud, and when it fits.
Cloud servers are virtual machines billed while they run. What drives the cost for game servers, why shutdown matters, and on demand vs reserved vs spot.
Network egress is data leaving a provider's network. What drives it in multiplayer games, why cloud meters it and bare metal includes it, how to estimate it.
A noisy neighbor is a tenant whose load can slow yours. How container limits keep it out of your match, what avoiding it costs, why big fleets share.
A vCPU is usually one hardware thread, not a full core. Why vCPUs differ, how many a game server needs, and how per-minute and reserved pricing compare.
Hosting & Orchestration
How game servers are packaged, started, scaled and shut down.
Agones runs dedicated game servers on Kubernetes. How it protects live matches, what running it involves, and what KRAFTON had in place for PUBG.
Auto scaling adds and removes game server capacity with demand. What a policy decides, why scaling down is the costly half, and how fast starts change it.
A cold start is the wait for a new game server's container to be ready for a match. What happens in it, why start times disagree, and what warm pools cost.
Colocation means sharing infrastructure. Why sharing a host with other studios' game servers is safe, why it costs less, and when owning racks pays.
A container runs one game server for one match. What it isolates, why port mapping trips up first deploys, where logs and results go, and how to recycle.
A dedicated server runs the match with no player on it. How it differs from a listen server, how each engine builds one, and why scope drives the cost.
A deployment is one running game server build. What it contains, how it starts and stops, how builds update without downtime, and what drives its cost.
Docker builds the image your game server ships as. How it differs from a VM, four build traps to catch locally, and why WebAssembly can't replace it yet.
Fleet management decides how many hosts your game servers run on, and where. Reserved fleets vs on-demand, how it pairs with orchestration.
Game server hosting is four layers: machines, placement, orchestration, operations. Where servers run, reserved vs on-demand, and when to build your own.
How orchestration runs each match, from placement to teardown. Regional hubs vs per-match placement, fixed vs cloud capacity, and why most games go hybrid.
A headless server runs the game with no rendering or audio. What breaks when you strip them, why it fits in a container, and how it shapes hosting cost.
Hybrid orchestration runs game servers on fixed hosts and cloud under one control plane. How spillover works, how to size the baseline, and vs multi-cloud.
Kubernetes automates containers for web services. Where it fits in a multiplayer backend, and what self-hosting game servers takes per region and release.
A multi-room game server runs many matches in one process. Which netcodes support it, how backfill fills rooms, where to place it, what a match costs.
What "serverless" means for game servers: functions, per-match servers or relays. The architecture's parts, what makes it work, and when it doesn't fit.
A warm pool holds game servers ready before matches ask. What it holds, how to size it, and why on-demand turns it into a cost decision, not a requirement.
Grouping Players (Matchmaking, Server Browser, Lobbies, Sessions)
Everything related to grouping players ahead of a match, and keeping seats filled in a session.
Backfill puts new players into matches in progress. How the server requests them, what backfill is used for, the rules it runs, and when to skip it.
A matchmaker groups players into matches: how it's structured, the rules it uses, what running one takes, and how fill rate drives hosting costs.
Server allocation binds a formed match to a game server. Where it sits between matchmaker and players, claiming vs starting a server, and why it fails.
Server regions split your matchmaking queue and fleet by geography. What that costs, how per-match city placement works, and when a fixed region pays.
Session fill rate is the share of player seats in use over a match. How to measure it, where it drops, genre benchmarks, and why bots hide empty seats.
Skill-based matchmaking groups players by ability. Where it sits in a matchmaker, what a game like Call of Duty's data shows, how tight a band to set, and when it fits.
Netcode
Synchronization, data transfer, rollback and overall impact on player's online experience.
Client-side prediction shows a player's input before the server confirms it. What gets predicted, how reconciliation replays inputs, why latency matters.
Desync is when a player's view of a match stops matching the server's. The two kinds, why tick rate isn't the fix, and when desync is a cheat.
Why host migration keeps interrupting matches, what has to move when the host leaves, what it costs to build, and how dedicated servers remove it.
Interpolation draws other players between server updates. What it costs, why Source defaults to 100 ms, and when games predict instead of interpolating.
Netcode keeps a multiplayer match consistent. The seven parts, what each one trades away, and why "bad netcode" is often a different layer.
Ping is the round-trip time between a player and a server. How it shapes netcode, why the gap between players matters more, and what developers control.
Rollback netcode predicts, rewinds and replays. Why latency sets the size of every correction, why games mix in input delay, and what it costs to build.
Rubberbanding is the snap-back when a server corrects a mispredicted position. What causes it, why it isn't just lag, and how to cut it server-side.
Tick rate is how many times per second a game server recalculates the state of the match, measured in hertz.
Network Architecture
How clients, servers and relays connect.
Client-server architecture routes every player through one server. How the game loop works, why clients predict and interpolate, and what the hub costs.
Crossplay puts console, PC and mobile players in one match. Why peer-to-peer struggles with it, why versions must match, and how to keep the queue whole.
Peer-to-peer multiplayer has no hosting bill. What it costs instead: failed connections, host cheating, matches lost when the host quits, and support.
A relay server routes players around NAT without hosting the match. What it fixes, what stays peer-to-peer, what free relays cost, and when to move on.
Players
Everything related to functions of keeping a game live, operating it, and ensuring a fair and excellent online experience for players.
