Using the incremental approach we achieved better querying times than the previous implementations (that had standard data structures for querying which were not specifically designed for DQIs). We also consumed less memory as we created the data structures whenever they were needed. The new approach enabled us to give preview bar, histogram, and the tight coupling information to the user without making any additional queries or spending additional processing times. Demanding a large main memory still remained as a secondary problem. We saw that there are problems in the selection time before we reached to the memory limits of our architecture (more than 1 second generally annoys the user). The selection times were mostly equivalent to 1 second. Hence, memory still remained as a secondary problem in DQI. As r increases, terms that contain the r factor become more significant. The starfield display times were significant for huge r's and our approach gains its power from the incremental starfield display updates (but huge jumps in range sliders can still cause higher display update times).
The average deviation for the selection times was smaller than we expected for the random set of experiments. The setup times were also acceptable as the disk input caused fluctuations in setup times. The querying time estimates were less accurate than we expected but were again acceptable. The reason for this was, we made high precision measurements and the system we used has lower precision settings than we needed.