Results of Simulation V8 show a clear improvement of the PeerC1 algorithm to its predecessor and an slight improvement over the "normal" BitTorrent algorithm. It is open to discussion what are the reasons for this improvement and more important how and to which extend the different adaptions of the V7 algorithm have transformed them self in effect in the latest simulation.
Adaptions over the previous algorithm
In the V8 code of the PeerC1 algorithm the main changes have been the following ones:
Adding of a random sleeping period before spawning of nodes ( Issue #57 )
Improved Peer discovery by dropping uninteresting peers and requiring the Tracker for new peers ( Issue #55 )
Setting the OU Slot length back to 30 ticks like in the normal BitTorrent protocol ( Issue #60 )
The different adaptions of the algorithm mentioned above have been done in order to improve the algorithm in the following way:
SleepTime: Because of the discrete ( tick orientated ) Simulations nodes have been indirectly synchronized. Adding a random sleep to nodes, breaks this synchronizations, or at least groups them together in groups depending on their initial sleep time.
One of the goals of the PeerC1 algorithm is to improve peer discovery by allocating OU slots dynamically. This only makes sense if the peer has the possibility to OU discover new nodes. For that reason uninterested peers ought to be dropped and the peer list filled with new peers. This is done by dropping peers that have no interest in our data and vise verso. In order to no DDOS the Tracker, peers have to limit the Peer List update frequency which witch to equiere new peers from the Tracker.
Adapting the number of max TFT/OU Slots was done because in the Simulation V7 results it was clear that too much bandwidth was allocated to OU Slots and not enough to TFT. This was the case because with a piece availability of about 20% and higher, four OU and six to eight TFT slots would have been allocated. As bandwidth is shared between OU/TFT based on the number of slots, OU would get up to 40% of the total upload bandwidth even though there are more than enough peers to cooperate. Resulting in a too low usage of the TFT principle.
Context
Results of Simulation V8 show a clear improvement of the PeerC1 algorithm to its predecessor and an slight improvement over the "normal" BitTorrent algorithm. It is open to discussion what are the reasons for this improvement and more important how and to which extend the different adaptions of the V7 algorithm have transformed them self in effect in the latest simulation.
Adaptions over the previous algorithm
In the V8 code of the PeerC1 algorithm the main changes have been the following ones:
Intended improvements
The different adaptions of the algorithm mentioned above have been done in order to improve the algorithm in the following way: