In the algorithm in Figure 5, we assume that read and write operations are processed following standard 2PL and ROWA protocols.
Given x [element of] D which is replicated at [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], 2PL and ROWA guarantee that any two conflicting transactions accessing x will access copies of x from at least one common site.
Transactions that update x always update all copies of x in [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], including [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], following 2PL and ROWA protocols.
2PL and ROWA, therefore, guarantee that any execution of Q, [T.
The ROWA protocol can also be replaced by other protocols such as quorum consensus, missing-writing, or virtual partition protocols.
We compare it with simple ROWA and 2PC protocols and focus on situations in which the MVA protocol performs differently.
The reread overhead causes the raw two-view access protocol performance to be worse than the ROWA protocol when the read operation is over approximately 95%.
In order to satisfy the "write all" requirement of the ROWA paradigm, it must be possible to locate all copies of a given page.
Because of the ROWA protocol, transactions executing under an avoidance-based scheme can always read any page copy that is cached at their local client.
When a transaction wishes to update a cached page copy, the server must be informed of this write intention sometime prior to transaction commit so that it can implement the ROWA protocol.
This is necessary to maintain the ROWA semantic guarantees that provide the basis for the correctness of avoidance-based algorithms.