How does Berkeley realize the transaction?

Question

I heard the Berkeley DB supports transaction feature, and I'd like to know more about it. In my mind, there are two cases about transaction.
I. One case is to do the transaction on one table
II. Another case is to do the transaction on multiple tables, which means those tables have relations, and transaction needs be bound on those tables
Which case is supported by the Berkeley database?
Please help me to understand it.

Answer 1

Hi Ashok,
Let me clarify my example firstly. In that example, there are three threads (thread-1, thread-2, and thread-3), they are trying to access the database with transaction.
For thread-1, it starts a transaction, and updates A, but because of the thread (task) schedule, thread-2 gets the CPU and starts another transaction. Similarly, when thread-2 updates B, thread-3 gets the CPU and starts the third transaction. After thread-3 commits/rollbacks the third transaction, thread-2 acquires the CPU again, and continues its transaction, then thread-1 acquires the CPU, and continues its transaction also.
Of course, for a database, it can be designed to DO NOT allow transaction-crossing, which means when a thread is doing the transaction on DB-A, other threads cannot do the transaction on DB-A, until the former thread finished the transaction operation. However, it is a technology issue, not a requirement issue. From the technology perspective, a database can be designed to DO allow transaction-crossing, which means when a thread is doing the transaction on DB-A, other threads still can do other transactions on DB-A. Certainly, in order to realize the transaction-crossing, it needs more efforts, includes well defined data structures, the algorithm to correlate multiple transactions, the policy to store the data during a transaction, etc... So, I'd like to collect some feedbacks from people who are using or designing the database, and make a decision if the transaction-crossing function is necessary for a database product. If most of designers and users do not think the transaction-crossing function is useful, then I am over considering it.
Kind Regards,
Kevin.
BTW, in my example, symbol "x--------> y" means thread-switching (from thread x to y), symbol ".........." is used to make the text be aligned well, it has not too much meaning, symbol "x<----------y" means thread-switching (from thread y to x)