Also, the choice between varbinary(4) (which should really be binary(4)) and int comes down to deciding between being able to natively bitmask IPs with subnet from within SQL Server or not. The reason I chose int regardless of the shift into negative is that I can write a stored procedure to retrieve all IPs of a given subnet, which could not be possible with binary(4) and would require me to retrieve the full list and then bitmask filter it in the application layer.