Sorry about that, had to finish up quickly. I will try to clarify my view, and sorry again if this wall of text is too long. I just wanted to stress that Consis

Sorry about that, had to finish up quickly. I will try to clarify my view, and sorry again if this wall of text is too long. I just wanted to stress that Consistency is about data itself, while Isolation is about dataset acquisition (the same goes for Atomicity). My view is that Consistency should only be mentioned when talking about data. It is much like method argument type checking in type-safe languages: if you get float passed in it WILL contain a valid number and you do not need any additional explicit checking for that. Consistency is data validation. Isolation, on the other hand, is a form of concurrency control. Once you have shared data between concurrent threads, care must be taken. SQL brings this to a new level by having only shared data between threads (transactions). Since performance is directly related to concurrency control techniques, some compromise must be taken. The article is just about those compromises. Phantom reads can be viewed as a manifestation of holding and releasing locks (semaphores) during thread execution. I believe that ACID properties are all orthogonal and databases should be examined where/when/how those properties are violated, e.g. Isolation is not always ensured because of the mentioned MVCC. Atomicity: data is treated in atomic datasets; Consistency: data validation; Isolation: data immutability during transaction; Durability: data retention (and as a consequence Consistency) over prolonged periods of time. But that is my own humble opinion.