GamingGaming GlossaryPart 3 of 8
Multiplayer Glossary
64 terms from the Gaming book's Multiplayer part, each defined in a line and linked to the chapter that explains it.
- Agones: an open-source library for hosting, running and scaling dedicated game servers on Kubernetes, with GameServer and Fleet resources and an SDK the server calls (Game Servers & Hosting).
- Allocation: handing a forming match a server that is already Ready and returning its address and port; idempotent (keyed by the match ID), atomic and placed by latency (Game Servers & Hosting).
- Authoritative server: a server that runs the real simulation; clients send inputs and receive state, and when they disagree the server wins (Netcode Fundamentals).
- Backfill: matching players into a match that already exists, such as one a player left, through a ticket that describes the open slots (Matchmaking & Ranked).
- Block: a one-directional edge that needs no consent and overrides the rest of the social graph: it removes a friendship, hides presence both ways and stops whispers, invites and voice, enforced by every service (Parties, Presence, Chat & Voice).
- Client-side prediction: the client running the same movement code as the server on its own inputs and showing the result at once, instead of waiting a round trip (Netcode Fundamentals).
- Connected realms: realms joined to share one population, the old realms kept as virtual realms running on a host realm's hardware (Game Servers & Hosting).
- Dedicated server, listen server: a server with no local players, against one where a player's machine hosts the match, has zero latency to the authority and ends the match when it leaves (Netcode Fundamentals).
- Desync, first divergence: machines whose simulations no longer match, found by comparing state checksums per frame or turn; hierarchical state logs are diffed to find the first divergence, because everything after it is noise (Lockstep, Rollback & Determinism).
- Deterministic lockstep: every machine running the whole simulation in turns from identical commands, a turn executing only when every player's commands have arrived, at the speed of the slowest connection (Lockstep, Rollback & Determinism).
- Drain: telling a server to finish its match and shut itself down rather than killing it; scale-down and updates never touch an Allocated server (Game Servers & Hosting).
- Elo, K-factor: one rating per player, moved after each game by K times the difference between the actual and expected score; FIDE uses K of 40, 20 or 10 by experience and rating (Matchmaking & Ranked).
- Entity interpolation: drawing other players a little in the past, between the two server states that bracket the render time; the interpolation delay is the floor on how old every enemy on screen is, and extrapolation covers only a briefly late packet (Netcode Fundamentals).
- Fan-out (presence): telling every online friend allowed to see it when a player's presence changes; its cost is concurrency times change rate times recipients, cut by coalescing changes (Parties, Presence, Chat & Voice).
- Fleet, ready buffer: a set of identical servers (one build, one location) kept warm for allocation, and the number held Ready at all times so matches don't wait for a server to start (Game Servers & Hosting).
- Frame delay: a small input delay in a rollback game that reduces the number and length of rollbacks; GGPO's guidance is that more than one frame is noticeable in fighting games (Lockstep, Rollback & Determinism).
- Friendship: a mutual, consented edge stored once per pair and hung off the account, so one friends list covers a publisher's games (Parties, Presence, Chat & Voice).
- Games per core: the capacity unit of a game server host at a tick rate; VALORANT targeted three 128-tick games per core, 2.34 ms each, with 10% of the frame reserved for the host (Game Servers & Hosting).
- GGPO: the MIT-licensed SDK that pioneered rollback networking in peer-to-peer games in 2009, built on save, load and advance callbacks the game implements (Lockstep, Rollback & Determinism).
- Glicko-2: a rating with a rating deviation (RD) and a volatility, updated per rating period; the deviation shrinks as evidence arrives and grows while a player is inactive (Matchmaking & Ranked).
- Head-of-line blocking: a lost packet holding up everything behind it until it's retransmitted, as in TCP; the reason real-time games run over UDP and build only the reliability they need (Netcode Fundamentals).
- Input recording, keyframe: a replay stored as the starting state or seed plus every input, tied to the exact build, with periodic state snapshots so playback can seek (Lockstep, Rollback & Determinism).
- Interest management, relevancy: deciding which entities each client receives and how often; it saves bandwidth and server CPU, and what a client is never sent a wallhack can't show (Netcode Fundamentals).
- Iris, Replication Graph: Unreal's opt-in replication system (Experimental in the 5.8 docs), and the earlier system of persistent nodes and grid cells built because Fortnite's 100 players and about 50,000 replicated actors outgrew per-actor checks (Netcode Fundamentals).
- Lag compensation, rewind: the server reconstructing hitboxes as the shooter saw them, judging the shot there and applying the result in the present, so a target can die after reaching cover on their own screen (Netcode Fundamentals).
- Maintenance window, downtime: a scheduled period when worlds are offline for changes that need them stopped, as opposed to data hotfixes and rolling restarts (Game Servers & Hosting).
- Match quality score: a proposed match's predicted closeness plus penalties for ping, party mismatch and role fit, accepted when it clears a threshold that falls as tickets age (Matchmaking & Ranked).
- Matchmaking rating (MMR): the hidden rating a game uses to make fair matches, kept separate from the visible rank and never shown to the player (Matchmaking & Ranked).
- Message classes: unreliable latest-wins state that is never retransmitted, reliable ordered events on their own channel, and inputs made reliable by repeating every unacknowledged one in each packet (Netcode Fundamentals).
- Mock game server: a lightweight fake that provisions, runs for a match length and reports a result without any gameplay, used to load test the platform at launch scale (Game Servers & Hosting).
- Open Match: Google's open-source matchmaking framework, whose parts (frontend, ticket pool, director, match function, evaluator, assignment) name what every matchmaker has (Matchmaking & Ranked).
- OpenSkill, Weng-Lin: a Bayesian rating method with closed-form updates for any number of teams, and the open-source libraries that implement it (Matchmaking & Ranked).
- Party: players who queue and play together as one ticket, with a leader and the states
open,queued,in_matchanddisbanded; any change while queued cancels the ticket (Parties, Presence, Chat & Voice). - Party MMR: the one number a party matches on, such as an average, a mean weighted toward the highest-rated member, or an average plus a party bonus (Matchmaking & Ranked).
- Peeker's advantage: the extra time a player moving around a corner has to react over the player holding the angle, from round trip and buffering; Riot measured about 141 ms in VALORANT and cut it to about 71 ms (Netcode Fundamentals).
- Placement (ranked): setting a new or returning player's rank from every prior signal, on the low end of where the system thinks they belong, then letting it converge (Matchmaking & Ranked).
- Placement (server): choosing the location that hosts a match from client-measured latency (ping beacons), minimizing the worst player's ping and capping the spread (Game Servers & Hosting).
- Prediction window: how far rollback may run ahead on predicted inputs (8 frames in GGPO) before it stalls; one rendered frame must be able to resimulate all of it (Lockstep, Rollback & Determinism).
- Presence, rich presence: whether a friend is online, derived from live sessions and held in memory, and the small, typed, localized detail of what they're doing (Parties, Presence, Chat & Voice).
- Queue, ticket: a pool of waiting requests plus the rules for matching them, and a request from one player or a whole party that can only match within its queue; a ticket moves from
searchingtoproposedtoassigned, or endscancelledorexpired(Matchmaking & Ranked; states in The Ticket Model). - Rank, LP, RR: the visible tier and division a player sees, with points (League Points, VALORANT's Rank Rating) to the next, moving in predictable steps toward MMR (Matchmaking & Ranked).
- Reconciliation: the client taking the server's state as of the last input it processed (by move number), dropping acknowledged inputs, replaying the rest and correcting smoothly if the result differs (Netcode Fundamentals).
- Relay: a server that forwards packets between players or to game servers without running the game, solving NAT traversal and hiding IP addresses, as Steam Datagram Relay does (Netcode Fundamentals).
- Rewind cap: the limit on how far back the server will rewind for hit registration, set from the latency distribution, with each client's claimed time checked against its measured ping (Netcode Fundamentals).
- Riot Direct: Riot's own network of routers in major internet hubs, peering directly with ISPs so player traffic avoids lowest-cost routing (Game Servers & Hosting).
- Role queue: a queue where each team needs one of each role and players list the roles they'll play; the scarce role sets everyone's wait (Matchmaking & Ranked).
- Rollback networking: applying local input at once, predicting remote input, and when a real input differs, loading the saved state and resimulating to the present within one frame (Lockstep, Rollback & Determinism).
- Server lifecycle: the states every orchestrator models: starting, Ready (in the pool), Reserved, Allocated (must not be deleted), Shutdown, and Unhealthy after missed health pings (Game Servers & Hosting).
- Server meshing: several servers simulating one shard, each holding authority over some entities and passing it on as they move (Game Servers & Hosting).
- Server Network Recording, Delta Checker: League's complete record of a match's inputs, settings and configuration, replayable on its deterministic server (recording replicated state, as Unreal's replays do, is the alternative that needs no determinism), and the testing tool from recorded inputs that the determinism work began with (Lockstep, Rollback & Determinism).
- Single shard: one persistent world for all players, split across servers by place rather than by copy, as in EVE Online (Game Servers & Hosting).
- Smurf: an experienced player on a new account, who stomps lower-skill matches while the rating catches up; most alternate accounts are made to play with friends of a different rank (Matchmaking & Ranked).
- Snapshot, delta compression: state sent as everything relevant to a client, or as the differences from the last snapshot the client acknowledged (Netcode Fundamentals).
- State synchronization: inputs up and state down, with the server simulating and clients predicting their own character; tolerant of imperfect determinism (Lockstep, Rollback & Determinism).
- Tick rate: how many times a second the server simulates and sends updates; higher ticks mean fresher state and less buffering for linearly more CPU, so a tick rate is also a hosting budget (Netcode Fundamentals).
- Time Dilation (TiDi): slowing the game clock on an overloaded server so every player there slows equally instead of the server falling behind (Game Servers & Hosting).
- TrueSkill, TrueSkill 2: Microsoft's Gaussian team rating updated by message passing on a factor graph, and its successor that adds experience, squads, kills and quitting as evidence (Matchmaking & Ranked).
- Voice token, voice channel: a short-lived credential your server signs for the voice vendor, naming the action, user, channel and expiry, and the channels it opens: positional (3D) per match, non-positional (2D) for teams and parties (Parties, Presence, Chat & Voice).
- WebTransport: HTTP/3 over QUIC in the browser, with unreliable datagrams for state and reliable streams for the rest (Netcode Fundamentals).
- Widening, expansion: relaxing a ticket's constraints on a schedule as it ages (FlexMatch calls the steps expansions), skill before latency, with a cap on how far it goes (Matchmaking & Ranked).
- World, realm (persistent shard): one copy of a persistent game world with its own population and database; not the same as a platform shard, which is a regional backend deployment (Game Servers & Hosting).
- Write-behind (character state): character state held in server memory and flushed to the database on a cadence, so a crash loses the time since the last flush; anything that moves value is written in its own transaction instead (Game Servers & Hosting).
- XMPP: an XML protocol for messaging and presence with rosters, which Riot's chat is built on and extends (Parties, Presence, Chat & Voice).
- Zoning, seamless handoff: simulating each region of a world on its own server, with a transfer (a loading screen, or an invisible handoff) when a player crosses between them (Game Servers & Hosting).