Closed mapa17 closed 12 years ago
Ok the error was the naive approach to have a moving average over the last two values.
The problem is that because the connection list is indirectly ordered in a way that new peers come at the end, the last average summations are done over peers with a very low piece availability and by doing so, lose all the previous knowledge.
Results of the Simulation V6 have shown that there is a bug in the piece Availability function. The calculated value is far to low for the number of pieces the different nodes have.