Capacity planning is the discipline of sizing a surveillance system so it actually handles its load — not just on day one, but as it grows. It spans several resources at once: storage capacity, storage write throughput, network bandwidth, server compute for recording and analytics, and client decode for viewing. The core insight is that these resources do not run out together; each hits its own limit at a different camera count, and the system breaks at whichever limit comes first.
The method is to compute each resource's demand from the camera plan and find the binding constraint. Storage has two separate budgets that are easy to conflate: total capacity (terabytes, from bitrate × time × cameras × recording factor) and sustained write throughput (megabytes per second, multiplied by the RAID write penalty) — an array can have plenty of terabytes yet fail to absorb the write stream. Recording servers are sized by sustained throughput rather than raw camera count; a server might handle on the order of a few gigabits per second of write, which is hundreds of cameras, before throughput, not "camera count", caps it.
The pitfalls are planning one resource and ignoring the rest, and forgetting headroom. Sizing storage capacity while overlooking write throughput, or buying GPUs while ignoring decode, yields a system that stalls below its nameplate. Good capacity plans include 20–30% headroom, redundancy (RAID plus a spare, N+1 failover servers, UPS), and a growth path, so the system meets its retention and frame-rate targets under real, sustained load rather than only in the brochure.
