Some questions about end game mode and certain peerwire parameters

[ZeroCool26]

I'm currently evaluating end game mode using Omnet++ simulator and im trying to figure out which parameters to change and see what happens but first i'm trying to figure out what i expect to happen before i start measuring things.Here is what im trying to figure out:

1)The BitTorrent specification says that if we unchoke 4(default) peers which have the best upload rate this will maximize the download rate of the client. Let's consider that we enter end game mode and 5 peers instead of 4 are unchoked.The fifth peer will certainly have worse upload rate than the other 4.How do you expect this to affect the download rate and the average download time and why?Do you expect to have more duplicates?

2)Consider that we are in end game mode and we optimistically unchoke more than 1 peer.How do you expect this to affect average download time and duplicates and why?

3)As far as block size is concerned, download time increases as the block size decreases.As the block size increases the download time increases slightly for large blocks.This is known as the 'last block problem' which is solved by BitTorrent but it results in redundant data(duplicates).This problem occurs when the last block comes from a slow source.What if it comes from a fast source?In general do you think block size would be a good parameter to change in order to evaluate end game mode and why?

I would much appreciate your thoughts on these 3 topics.Thanks in advance.

[Switeck]

1 and 2...sounds more like you're talking about your client's UPLOAD. There is only 1 optimistic unchoke upload slot per torrent. uTorrent attempts to download from lots of seeds/peers at once even when NOT in end game mode, certainly way more than 4.

3."As far as block size is concerned, download time increases as the block size decreases."

Huh? As far as I know, download time increases with piece size. If a torrent was made with 4 MB pieces, getting completed pieces hashed can be problematic with seeds/peers averaging less than 2 KB/sec upload to you each.

[ZeroCool26]

Let me rephrase my questions:

1)How does an increase of downloaders(more than 4) affect download time and duplicates in end game mode and why?

2)How does an increase of optimistically unchoked peers(more than 1) affect download time and duplicates in end game mode and why?

3)How does an increase of block size affect download time and duplicates in end game mode and why?

All i want to know is what you expect will happen in each of the above situations.I'm not talking about utorrent specifically.

[Switeck]

1)Even when there's ~20 peers on a torrent, endgame mode only occurs with the last couple pieces...which likely fewer than 4 downloaders/peers have. Duplicate data as a result is often low. Being connected to lots of seeds (vs few) also increases also typically increases duplicate data. If the seeds/peers are generally SLOW, I would expect duplicate data to be greater.

2)To my knowledge, ONLY BitTyrant has more than 1 optimistically unchoked peer per torrent...and then often only briefly! BitComet (and clones) and Transmission allow far more upload slots than they should, but I don't know if those are technically optimistically unchoked peers. I would expect having a high upload slots per torrent value (in uTorrent or any other client) would be effectively the same thing. The end result of too many upload slots depends on the average upload speed PER upload slot. If it's below ~1 KB/sec, those peers may SNUB the downloader and give it nothing. If it's between 1-3 KB/sec, they may intermittently upload to it slowly. If it's 3+ KB/sec, they will likely upload to the downloader at least at their average upload speed per upload slot...which could result in increasing amounts of duplicate data in end-game mode *IF* there's many peers doing this.

3)I'm not sure if block size is technically a BitTorrent protocol term...do you mean piece sizes (often 256 KB to 4 MB in size) or chunk size (typically 16 KB)?

A large piece size SHOULD take longer to download, following an exponential curve, with 4 MB piece size taking potentially 100 times longer than 256 KB piece size. This is due to unreliable peers and seeds which often only give 15-30 second bursts of upload until changing to a different peer to upload to...or they just plain disconnect. Because a disconnected (or given up) peer often fails to finish at least its last 16 KB, duplicate data can be considerable.