Multi-Room Game Server
A multi-room game server is one server process that hosts several matches at once, each in its own room with its own state and players. The process's fixed costs, such as the engine and loaded game data, are paid once and shared across every room, so each match needs only a fraction of a vCPU. It suits games with light simulations, such as card, turn-based, casual and mobile games, and gives them server authority at a cost close to a relay's.
Also called
multi-room server, multi-session server
,
What Is the Difference Between a Multi-Room and a Multi-Process Server?
"Does each match get a game server?" is a common first question in multiplayer. It doesn't have to. Several matches can share one machine in two ways:
Multi-process. Each match gets its own server process, and several processes run side by side on one host, each on its own port.
Multi-room. One process hosts many matches at once, each in a room with its own state and players.
The difference is what a new match receives: a new server, or a free room in one already running. Inside it, the game's own code keeps rooms apart.
Which Engines and Netcodes Can Run Several Matches in One Process?
Many room-based frameworks do it by design. Unreal points the other way.
Colyseus. "One room class can spawn many instances (e.g., multiple matches of the same game type)" (Colyseus documentation).
Nakama. Runs many authoritative matches per node, each with its own loop (Heroic Labs documentation).
Mirror (Unity). Its Multiple Additive Scenes example gives each match a subscene with its own physics scene (Mirror documentation).
Unreal. Its dedicated server usually runs one map and game mode per process; the common advice is several processes per machine (Epic Developer Community forums, 2016). Unreal games mostly scale by process rather than by room.
The choice is easier early: converting a single-match server means every piece of shared state, from singletons to the physics world, has to become per room.
How Do Players Get Into a Room?
Running servers list their open slots, and players reach them in one of two ways.
A server list. Players pick a running server with open slots, which suits persistent and community games. See server browser.
The matchmaker. It takes a group of players and decides: backfill them into an open room, or deploy a new server when no room fits. Joining feels like any matchmade game.
Either way, you scale by the room. To retire a server, stop sending players to it and let its rooms finish (containers covers recycling).
Where Should a Multi-Room Server Run?
A multi-room server's location is chosen once, when it starts, and every room plays from there. Per-match placement instead picks a location for each match's players (game server orchestration).
Rooms pay off when full, so multi-room servers gather where players are dense. Spread them wider and each fills more slowly; keep them few and distant players connect from farther away.
Turn-based, card, puzzle and casual games, and lobbies, tolerate a few extra milliseconds. Modes where latency decides the result keep a server of their own. Many games run both.
A word from our sponsor (ourselves!)
Launch day is a bad time to discover an orchestrator's edge cases. Edgegap's automated orchestration has run 135 million game server deployments since February 2019 across its distributed, multi-cloud network, and today deploys on behalf of thousands of games and multiple millions of players daily.
What Does a Match Cost on a Multi-Room Server?
A server process carries fixed costs before any match starts: the engine, loaded game data, baseline memory. Multi-room shares them across rooms, which is how a dedicated server gets close to a relay's cost while still holding the match.
An estimate, for a card game with 15-minute matches:
Server. 2 vCPUs hosting 40 rooms, so each match uses 0.05 vCPU.
Hour. 2 vCPUs × 60 minutes × $0.00115 = $0.138 on Edge Cloud (platform data, 18 September 2026).
Match. At 85% of rooms busy, that hour carries 136 matches: about $0.001 per match.
Then comes egress: compute stays flat per server, while egress grows with every player. At Edge Cloud's $0.10 per GB, a 2 vCPU server's egress equals its compute at about 150 players sending 20 kbps each, or about 3,000 players at 1 kbps. The denser the rooms, the more egress becomes the largest line on the bill.
That makes Private Fleet worth weighing: reserved hosts in 58 locations with egress included, and automatic overflow to Edge Cloud when they're full. The trade is reach, since Edge Cloud runs in 615+ locations. A dense baseline on Private Fleet, with Edge Cloud for peaks and farther regions, gets both.
Cap rooms per process at what a profiled build measured, and recycle servers on a schedule. Learn how fill rate affects your hosting costs.
Edgegap's Take (just our opinion, take it with a grain of salt!)
Server Authority at a Relay's Price
Teams have usually picked relays and peer-to-peer because a server per match cost too much, rarely because a relay made a better match. A multi-room server changes that sum: each match gets a room in a mini server, about 0.05 vCPU above, and the server still holds the state.
That is a large upgrade from a relay. The server decides what happened, every player is one hop from it, no one has a host's advantage, and the match survives anyone leaving. Bloody Bastards, a mobile game with 40 million downloads, moved from peer-to-peer to dedicated servers and a matchmaker, with "zero cheaters since" (case study, 2026).
It is the cost-effective choice when the server's job is light: turn-based rules, a card table, a result to referee. A heavy simulation earns a right-sized server of its own.
,
