"Quick seeding" (or, a new initial-seeding mechanism)

[Honeyfrog]

At present, there are two types of seeding: "Regular" seeding, which seeds rarest pieces (if desired by peers) as fast as possible regardless of peer retrading efficiency, and initial-seeding, which goes more slowly by tracking peer retrading efficiency.

I propose a "Quick seeding" mechanism whose purpose is to create a "good" distributed copy as fast as possible, paying some (but not exclusive) deference to peer retrading efficiency.

The way it works:

1) For X time units, seed "regularly". Track peer re-trade efficiency but don't act upon it yet.

2) At the end of X, determine which peers were most efficient at re-trading while also capable of accepting the seeder's upload volume.

3) For Y time units, upload exclusively, at maximum rate, to those peers selected in 2), tracking efficiency if CPU permits but not acting upon it. (Pieces still incompletely sent in phase 1 will continue to be sent.)

4) Repeat.

"Quick seeding" will be considered a type of super-seeding in that it won't send duplicate pieces until a complete set has already been sent.

X and Y could be variably dependant upon the number of peers, percentage completion of the distributed copy, and/or upload capacity utilization of the seeder.

In most instances, newly created and posted torrents acquire a steadily increasing number of peers until creation of secondary seeds. Quick seeding will most closely approximate regular seeding when there are very few peers, and most closely resemble initial-seeding when there are many peers. Initial leeches will be gradually cut off in preference to an increasing stable of fast re-traders.

[Bioret]

this is what im talking about