Infrastructure planning
Size game and website servers from measured safe capacity
A memory specification does not tell you how many players a modded world can support or how many requests a website can serve at acceptable latency. A useful capacity estimate begins with a test of the workload you intend to run.
Write down what safe capacity meant in the test
Record software and game versions, mods, map or content, hardware, test duration, request or player mix, acceptable response time or tick behavior, and the bottleneck observed. A result from a quiet map is not automatically valid for a busy event.
The calculator expects the same load unit in peak demand and per-server capacity. Players, simultaneous matches and requests per second are not interchangeable. Provider minimum specifications can be useful prerequisites; they are not a benchmark of your actual workload.
Count working servers before adding a reserve
Suppose a measured server supports 40 simultaneous players under your stated performance criterion. A peak of 120 with 25% headroom gives a target load of 150. Dividing by 40 gives 3.75, so four working servers are needed.
Adding one explicit reserve makes five provisioned servers. At an illustrative USD 30 per server per month, the scenario costs USD 150 per month. A target just over an exact capacity boundary can require another whole server even when the increase looks small.
A reserve only helps if it can actually take the work
A counted reserve does not establish deployment speed, state replication, networking, placement across failure domains or spare capacity in a cloud region. Those need a tested operational design.
Storage, backups, bandwidth, egress, moderation, support, licenses and human administration are outside the simple per-server price. Record them separately before calling a community or hosted product profitable.
Retest the assumptions that changed
A new modpack, game release, content change or application feature may invalidate the benchmark. Retain the old measurement as evidence of an old configuration rather than silently treating it as current.
Use the canvas to keep the capacity result beside a recurring-cost or revenue scenario. Inputs remain independent in this release; a value is not automatically transferred between widgets or validated for a different workload.
Authored 2026-10-29. Arithmetic examples are covered by local regression tests. New workflow guides remain outside the indexing allowlist pending editorial and applicable specialist review. Examples use the assumptions stated here; authority guidance can change.