In the context of parallel computing, it has been said (quote?) that acquiring exclusive locks in opposite order will eventually lead to deadlock conditions. It has been suggested (source?) that the generic antidote for this (still within the context of the lock paradigms) is to acquire locks always in the same order. More so, within the context of Two-Phase Locking strategies, it is also demanded that transactions acquire all the locks they will need (Growth phase) before releasing any of the acquired locks (Release phase).
Question: Is it the DAG the more general abstract hierarchy that establishes a lock acquisition order without potential deadlock conflicts? Or it is a tree of locks? does the order of acquisition matter? (top-to-bottom versus bottom-to-top)