A little nitpick about > To those who still tend to use int4 in surrogate PKs, I have a question. Consider a table with 1 billion rows, with two columns – an i

A little nitpick about > To those who still tend to use int4 in surrogate PKs, I have a question. Consider a table with 1 billion rows, with two columns – an integer and a timestamp. Will you see the difference in size between the two versions of the table Wouldn't the important thing be an index size, not a table size? Table size already has 23-byte header + alignment padding. So 4 byte difference doesn't do much for table size. But fitting more of an index into memory could have some benefits. An index entry has 8-byte header. Secondly, 1 billion rows (used in the example) are far too close to the maximum for int4 for comfort. Great article nonetheless.