Analyzing the dimension-based histograms generated from all the databases, we remark a relation between the database generated from the analyzed program, and the program itself. Let us take the satisfiability test for example.
The program Ocean.f in the PerfectClub database: Observing
the distribution based on the numbers of variables of constraint
systems in 6_fig:PerfectClub_database_100_distribution we can
see a peak corresponding to dimension of 
, that shows the effect
of ocean.f. 6_fig:SPEC95_database_100_distribution,
6_fig:PerfectClub_database_filter_distribution and
6_fig:SPEC95_database_filter_distribution show distribution
of the other databases in PerfectClub and SPEC95, without explanation
of the peaks. The phenomenon is due to the complex structure of the
ocean's main module with 
lines of code, where a great number of
constraint systems are to be manipulated, and average sizes of these
constraint systems are much larger than the others because of the
large number of global variables visible and modified at the main
level.
In our experimental sections, in order to reduce the number of graphs and tables in this dissertation, we only present the results of PerfectClub or SPEC95, which are direct comparisons of implementations and characteristics of databases, and do not consider all the criteria. Any difference between results of the two benchmarks is revealed if it does exist. The full set of experimental results can be found at POLYBENCH's website [QUE].
