I am analysing algorithm performance, which has random nature.
Since its for security application the more random the better, which
will hide small dependencies under.

What i do is comparison its structure(it results in directed graph with
each vertex having one outer edge) to random mapping.

Now run tests, and look for graphs with number of vertex 128,
create random mapping graph, and graph for my algorithm and checks
number of cycles of different length and so, the number of different
graph is 128^128, what is my problem know that for number of test T=10000,
the results for both doesn't converge and oscillate.

Could you tell me how many test need to be run to get more convergenous
results? or point some source where such info can be got.