All you're doing is picking random =subsequences= from the same permutation of the original data.
You have some good points in your reply. I am very much aware of this non-random behavior you point out for the "static random-value column" approach but at least it is fast, which is a requirement. :-( However, if the life time of the individual rows are short, the behaviour is, luckily, sufficiently random for my specific purpose.
I furthermore realize that the only way to get truly random samples is to ORDER BY random(), but this is an unacceptable slow method for large data sets. Even though it is not trivial at all, there ARE indeed algorithms out there [1,2] for picking random sub sets from a result set but these are (sadly) not implemented in postgresql.
References: [1] http://portal.acm.org/citation.cfm?id=304206 [2] http://compstat.chonbuk.ac.kr/Sisyphus/CurrentStudy/Sampling/vldb86.pdf ---------------------------(end of broadcast)--------------------------- TIP 3: Have you checked our extensive FAQ? http://www.postgresql.org/docs/faq
