Paper #: 3.18
Title: Free Transactions with Rio Vista

1. Problems
-Atomic transactions have typically taken a significant portion of time, with multiple disk I/O's. Because of the time required to guarantee data integrity, many applications and especially system processes have been reluctant to take advantage of transactions, relying instead on their own checkpoints.

2. New Idea and Strengths
-This paper proposes a system that is designed to take advantage of the new technology called Rio, a persistant file cache in memory. The system, Vista, takes advantage of the guarenteed persistance of Rio to store information.
-Vista saves on the redo end, by not storing a redo log (which would be used by the system in the event of a system crash) and by not bothering to periodically push the changes to the databse.
-Vista is tailored specifically to Rio, which gives it a big advantage over other general transaction management programs.
-As the authors mention no less than four thousand times, Vista is only 720 lines of code, which apparently is some kind of miracle for a program that does what it does.

3. Weaknesses and Extensions
-The biggest weakness is that this is only applicable in a system with a Rio file cache. I don't expect Vista is portable to other systems.
-The biggest advantage Vista gives is in situations where the size of the transctions are very small. This makes sense in their ideal situation of never having to write the file cache to disk. However, in high volume situations it would be interesting to see how Rio/Vista performs. According to their graphs, it still provides an advantage over traditional transaction systems, but the advantage is significantly less than 2000 times at high levels of traffic. The biggest difference is that perhaps Rio/Vista will indeed be a successful choice for transactions on a system level, or at high levels with the Vista/Rio only combination.