Sizing
How much database will you need?
Describe your metering population and this estimates the SyncGrid database size — the time-series data and the runtime attributes algorithms write. Every storage number behind it is measured on a real Postgres, not assumed.
Estimate
Required database size
—
| Group | Assets | Sources | Samples/day | Algos/day | Retention | Time-series | Runtime attrs | Total |
|---|---|---|---|---|---|---|---|---|
| All groups | ||||||||
What is in the numbers
- Time-series stored as compressed whole-day chunks, one row per asset, source and day — including the primary-key index and Postgres row overhead.
- Data per sample picks how much each reading carries, per source group: value only is the value, its status and its insert time — a plain load profile, and what nearly every head-end sends, so it is the default; value + raw register adds the meter's raw cumulative register and its raw status word; extended adds two auxiliary values per sample, which carry extended events.
- Measured on a realistic synthetic profile at full day coverage, not modelled.
- Runtime attributes are one more time series per asset, carrying the three results per asset an average algorithm run writes (runs write one to five). Measured the same way as meter data, plus 20 % for the bookkeeping a runtime attribute needs and a meter value does not.
- Sizes are decimal (1 TB = 1000⁴ bytes), the way disks are sold.
Not included — read before you budget
- No Postgres high availability — a secondary/replica doubles the figure.
- No backup storage.
- No Postgres WAL.
- No audit log, alarms or other system data — likely an under-estimate, add around 10 %.
- No job result JSON. Algorithm results are kept under their own retention, which can be cut much shorter than the data retention, so they are sized separately.
- Virtual assets produce their own time-series; enter them as separate groups.
- Time-series numbers assume 0.1 precision, randomized sampling within limits and all values present. Real deployments differ — higher precision and duplicate values increase the size, while zeros (as in A-) decrease it.
What real data does to this
Data from real customers is typically 5-30 % lower than this estimate, because real series compress better than a synthetic profile — but it depends entirely on the actual data, so treat the range as a direction, not a discount.