FPABench measures how fast planning software loads data, recalculates a plan and answers reports. Every system runs the same P&L model on the same data. FPABench is made by Farseer, which builds a planning platform for finance teams. RAMA is the calculation engine inside Farseer and one of the entries. It follows the same rules as the others. About says more.
In plain words: finance teams build budgets and forecasts in planning software. When a number changes, every total that depends on it must update, and reports must open fast. FPABench makes up one realistic company and gives its data to every system. Then it times the everyday tasks: loading actuals, typing into the budget, opening reports and changing the plan. Why: vendors get one standard way to measure their system, and buyers get a fair way to compare tools.
Read this first. All benchmarks are wrong in some way, and this one is too. DuckDB and ClickHouse are databases, and they have no way to recalculate only what a write changes. After each write they rebuild every table that depends on it. So their write and model change times are not a like-for-like match with a planning engine. The limitations list every caveat.
The scale is the size of the dataset. Each size is named by about how many cells the model stores, so 10m holds about 10 million cells and 10b about 10 billion. Every system loads the same files. The data page lists what each size holds.
Lower is better. ×1.00 is the fastest entry on a test. Tick one machine to compare the systems on it, or several to compare machines too, as in ClickBench. Every row shown is scored against the fastest row shown. The method's reference machine is a Google Cloud n2d-standard-4, and the published rankings use it alone. No system is tuned or limited to fewer threads. A failed test counts as much slower than the slowest success. The method has the details.
Detailed comparison
Untick a test to leave it out of the score. The tests page says what each one does.