what-when-how
In Depth Tutorials and Information
Table 6.1
TheTorrentsUsedforthePiecePopulation
Experiments.
Torrent Name
Pieces
Size (MB)
Leechers
Clients
a
KNOPPIX
b
4125
4325
169
10
FreeBSD
b
5699
1494
34
10
mandriva
b
2803
735
89
9
openSUSE
b
14805
3881
398
9
feisty
c
1387
727
65-120
20
openSUSE-2
c
14977
3926
100-150
18
PlanetLab-1
c
1497
784
0-340
340
PlanetLab-2
c
1497
784
0-390
390
a
Number of administered clients used to connect to the torrent. All
peers are administrated clients in PlanetLab experiment.
b
Snapshots of population taken.
c
Evolution of population monitored.
Figure 6.1 shows the snapshots of the piece populations for various real torrents.
All four appear to be normally distributed with mean values slightly less than half
the downloaders, indicating that they are in the transient stage. he least normally
distributed populations are Knoppix and openSUSE, which correspond to their
being the largest torrents. his larger torrent size results in peers having a limited
local view of which pieces are rarest, which leads to a distortion of the normal curve
towards some pieces having extra copies (the tails evident in Figure 6.1). he other
two populations are small enough that a peer's local view is nearly complete, result-
ing in a near perfect normal distribution.
Table 6.2 conirms our visual analysis of the width of the distributions. he
largest torrent (openSUSE) has a block variation that is an order of magnitude
larger than the others. he smallest block variation is for the FreeBSD torrent,
which also has the fewest number of peers.
6.2.1.1 Evolution of Piece Population
To further understand the dynamics of piece population in the different stages of
the torrent, we have also monitored torrents throughout their lifetime. It is worth
noting that such experiments can be difficult to conduct for real Internet torrents
because, in general, we do not know the exact start time of a torrent unless it is
launched by ourselves. Torrents launched by ourselves, however, are not necessarily
representative and the measurement results can be biased. hrough constant online
