When an application reads more input and writes more output data than necessary, the unnecessary reads and writes impact performance. You can expect improved performance from data-in-virtual because it reduces the amount of unnecessary I/O.
As an example of unnecessary I/O, consider the I/O performed by an interactive application that requires immediate access to several large data sets. The program knows that some of the data, although not all of it, will be accessed. However, the program does not know ahead of time which data will be accessed. A possible strategy for gaining immediate access to all the data is to read all the data ahead of time, reading each data set in its entirety insofar as this is possible. Once read into processor storage, the data can be accessed quickly. However, if only a small percentage of the data is likely to be accessed during any given period, the I/O performed on the unaccessed data is unnecessary.
Furthermore, if the application changes some data in main storage, it might not keep track of the changes. Therefore, to guarantee that all the changes are captured, the application must then write entire data sets back onto permanent storage even though only relatively few bytes are changed in the data sets.
Whenever such an application starts up, terminates, or accesses different data sets in an alternating manner, time is spent reading data that is not likely to be accessed. This time is essentially wasted, and the amount of it is proportional to the amount of unchanged data for which I/O is performed. Such applications are suitable candidates for a data-in-virtual implementation.