Transactions are just like database queries. In fact, the bank account example is the canonical example in most introductions to database transactions. We want to make sure that it appears that the transactions are happening in the exact order specified. This is the abstraction, whereas the implementation may be vastly different. When performing writes, a batch of transactions will change memory state, but this change is only final if ALL transactions are successfully completed. If one or more fails, they all abort and state is returned to what it was before any transaction was made.
Here, does isolation actually means no other transactions can observe writes before commit? If there is a program in another processor spin checking that specific memory address, it should see the written result before the transaction is committed.
This comment was marked helpful 0 times.
fangyihua
No one should see the change of the write before commit since everyone has to commit at the same time.
This comment was marked helpful 0 times.
danielk2
Just to add, transactional memory is effective when either the program structure is too complex for the programer to implement fine-grained locking or there is not much conflict so that there isn't much rolling back.
Transactions are just like database queries. In fact, the bank account example is the canonical example in most introductions to database transactions. We want to make sure that it appears that the transactions are happening in the exact order specified. This is the abstraction, whereas the implementation may be vastly different. When performing writes, a batch of transactions will change memory state, but this change is only final if ALL transactions are successfully completed. If one or more fails, they all abort and state is returned to what it was before any transaction was made.
For the more curious
This comment was marked helpful 0 times.
Here, does isolation actually means no other transactions can observe writes before commit? If there is a program in another processor spin checking that specific memory address, it should see the written result before the transaction is committed.
This comment was marked helpful 0 times.
No one should see the change of the write before commit since everyone has to commit at the same time.
This comment was marked helpful 0 times.
Just to add, transactional memory is effective when either the program structure is too complex for the programer to implement fine-grained locking or there is not much conflict so that there isn't much rolling back.
This comment was marked helpful 0 times.