Re: No heap lookups on index

As an Oracle internals person myself, I don't see how making a comparison between the specifics of Oracle's MVCC to PostgreSQL's MVCC is relevant to this discussion.

As does *MOST* other commercial databases, Oracle's storage manager performs an update-in-place whereas PostgreSQL's (for the most part) does not.  There are several ways to implement update-in-place, and Oracle has chosen their own rollback segment methodology which has issues that without tuning, are major hassles.  I'm not saying that one is better than the other in ALL cases, but I and many other Oracle consultants have tuned Oracle installations to eliminate the headaches others in this list have mentioned.  Any knowledgable Oracle person evaluating PostgreSQL that may be reading this list is just going to see it as a lot of anti-Oracle discussion with no basis in fact.

Regardless, there is NO WAY to perform an apples-to-apples comparison between the implementations, locking strategies, etc. as the MVCC implementations and goals are completely different.  In short, Oracle's implementation is not perfect; neither is ours.  Oracle's initial design (as a commercial database) is much different than PostgreSQL's (as a research database).

While I'm always game for looking at other implementations when designing and discussing new features for PostgreSQL, let's focus on making PostgreSQL better rather than spending time discussing unrealistic comparisons.

If we want to do a comparison on the how/why Oracle's index implementation is faster in the context of this situation and how we could make PostgreSQL's faster, let's stick to that.