Server Fleet

A server fleet is the pool of game server instances — running or on standby — that an orchestration platform manages on behalf of a studio, scaling up and down in response to player demand.

When a matchmaker groups players together, it doesn’t start a server from scratch — that would take too long. Instead, it requests a session from a fleet of pre-provisioned servers that are ready and waiting. The fleet absorbs incoming session requests, assigns players to available servers, starts new instances when capacity is running low, and terminates idle ones when demand falls.

What fleet management involves

Running a fleet is more than just keeping servers alive. A fleet manager needs to:

Maintain a buffer of warm, available sessions so matchmaking requests are fulfilled in milliseconds rather than the 30–60 seconds it takes to cold-start a container
Scale across regions — players in Asia need servers in Asia; a single-region fleet creates unacceptable latency for global games
Handle failures — if a node goes down, sessions need to be redistributed without players losing their match
Manage image versions — when you push a game update, the fleet needs to drain sessions on the old version, pull the new image, and make it available without downtime
Track capacity — know at any moment how many slots are available per region, per version, per node pool

Building this yourself with AWS GameLift, Agones, or raw cloud infrastructure is the work of a dedicated DevOps engineer. It’s the core of what a managed orchestration platform like Gameye provides as a service.

Fleet size and billing

Most studios don’t pay for a fleet of fixed size — they pay for sessions consumed. The fleet scales automatically: small overnight, large at peak times, and the studio is billed only for what runs. This is fundamentally different from reserving a fixed number of game servers, where you pay for idle capacity regardless of whether players are in sessions.

The economics of a well-managed fleet compound over time. A fleet that maintains only a small warm buffer (enough to handle typical demand spikes with a short lead time) costs far less than one that keeps a large standing inventory “just in case.” Getting this balance right — low idle cost, fast response to demand spikes — is one of the core engineering problems that orchestration platforms solve.

Fleet management in practice

When Nakama’s matchmaker finds a group of players, it calls Gameye’s Fleet Manager. The Fleet Manager checks available capacity in the player’s preferred region, allocates a session from the warm pool, returns the IP and port, and instructs the pool to spin up a replacement instance to maintain the buffer. The entire exchange takes milliseconds. Players join a server that was already running and waiting — they never experience the cold-start delay.

Back to Glossary