netdata/libnetdata/storage_number/README.md

Netdata storage number

Although netdata does all its calculations using long double, it stores all values using a custom-made 32-bit number.

This custom-made number can store in 29 bits values from -167772150000000.0 to 167772150000000.0 with a precision of 0.00001 (yes, it's a floating point number, meaning that higher integer values have less decimal precision) and 3 bits for flags.

This provides an extremely optimized memory footprint with just 0.0001% max accuracy loss.