Time-series storage sizing

How it is calculated

• A raw sample is taken as 16 bytes — timestamp, value and tag reference — before compression.
• Compression ratios are the ones industrial workloads actually achieve: about 6× for gzip, 11× for Gorilla-style delta-of-delta, 18× for aggressive columnar encoding on slow process tags.
• Deadband suppression removes samples before they are stored, so it scales the whole number down linearly.
• The edge buffer is sized on the uncompressed arrival rate over the outage window, because that is what a store-and-forward buffer actually has to hold.

Where this stops being true

Compression ratio is the number that swings the answer and it is entirely workload-dependent: a slowly-drifting temperature tag compresses twenty times, a vibration waveform barely twice, and one noisy analogue tag sampled at 1 Hz can outweigh a thousand well-behaved ones. The figures here are representative of process historian data and will mis-size an array built for high-frequency condition monitoring. Measure a week of your own tags before committing to hardware — this is a planning estimate, not a procurement specification.