Hi there,<BR><BR>The whole point of a transaction is that a set of operations all happen together or none of them at all.<BR><BR>I was wondering how this is implemented.<BR><BR>Surely, for this to be perfectly tight, the final commit ought to boil down to a single machine instruction otherwise, for example, someone could switch the power off in the middle of the series of instructions which commit the transaction and have only some of the database modified.<BR><BR>Having said that, surely its possible to have a single machine instruction only partially complete.<BR><BR>So, what I&#039;m asking is, do transactions in the real world really satisfy the ACID test regardless of any exceptional circumstances?<BR><BR>I.<BR>