Compressed bucket-ranges works fine for random date, but if the data are skewed in some way (e.g. normal-curve shaped), the bucket-sizes must be adjustable. An

Compressed bucket-ranges works fine for random date, but if the data are skewed in some way (e.g. normal-curve shaped), the bucket-sizes must be adjustable. An alternative method is to use one bucket into which all values below a limit (e.g. 20,000,000) are sorted as they arrive, and compress all the rest. When 10^6 values have arrived, transmit the bucket and then reuse the empty bucket repeatedly to sort the compressed values.