Controls which accumulator to use for SUM or AVG of floating-point numbers.
DBA permissions are not required to set this option. Can be set temporary for an individual connection or for the PUBLIC group. Takes effect immediately.
The small accumulator for floats and doubles is highly accurate for addends in the range of magnitudes 1e-20 to 1e20. It loses some accuracy outside of this range, but is still good enough for many applications. The small accumulator allows the optimizer to choose hash for faster performance more easily than the large accumulator. The large accumulator is highly accurate for all floats and doubles, but its size often precludes the use of hash optimization. The default is the small accumulator.