what-when-how
In Depth Tutorials and Information
contents larger than 100 MB. It also worth noting that 50% of nonvideo contents
are less than 20 M, whereas those small contents are very few in the video torrents.
Figure 6.26 shows the cumulative distribution of BT torrent size. his distribu-
tion is relevant to the popularity of different BT contents. We learn that although
the video torrents are mostly larger than nonvideo torrents, more than 90% tor-
rents have less than 100 peers (more than 95% torrents have less than 300 peers).
According to these basic properties, we know that the video torrents have poten-
tial to raise more inter-ISP traffic problems due to its large content size and torrent
size. In particular, if the peers of a video torrents are uniformly distributed among
ASes, it is more likely to generate heavy traffic through the backbone.
In order to understand such a challenge, we randomly select 8893 BT video
torrents, and collect the AS information of every peer in each torrent. his prob-
ing is based on the “whois” command on Linux system and most replies are from
“whois.cymru.com.” From Figure 6.27, the AS popularity of video BT peers its the
exponential distribution, that is, among all 2405 ASes in our measurement, most
of them have less than 10,000 peers in total. Based on our measurement results, we
also present the Top-10 ISPs/ASes with the most video BT peers in Table 6.3. his
result can also be regard as the challenge and the potential requirements of P2P
locality in these popular ASes.
We further investigate the AS distribution of different video torrents in Figure 6.28.
In this figure, 141 small video torrents and 39 big video torrent are selected. Each point
in the figure indicates the number of peers in the AS, and the values are all sorted in
descending order. We can learn that the AS distribution of the large torrents are more
uniform than that of small ones and involves more ASes.
Figures 6.29 and 6.30 show the ratio between AS cluster size and torrent size. In
Figure 6.29, we first observe that this ratio is quite high in small torrents; the largest
AS cluster can even reach to 30% of the torrent size. herefore, due to the small peer
population, these torrents already have some locality features in nature. Consequently,
the benefit of locality mechanism will also be limited by the total peer number.
In the case of large torrents, Figure 6.30 shows that although large AS cluster is
more likely to exist in the big torrents, its ratio to the torrent size is relatively very
low. In particular, the largest AS clusters only have less than 6% of peers in the AS.
Moreover, we find that the distribution of this ratio can be fitted by Mandelbrot-
Zipf distribution with
α
= 1.33 and
q
= 10. he MZipf distribution deines the
probability of accessing an object at rank
i
out of N available objects as:
p
(
i
) =
K
/(
i
+
q
)
α
, where
K
Σ
1
1 ( )
α
,
α
is the skewness factor, and
q
≥
0 is the plateau
factor.
q
is so called because it is the reason behind the plateau shape near to the left
part of the distribution.
hese measurement results indicate that the size of most AS clusters is quite
small. According to the definition of locality [20], we believe that a global locality
approach may not be our best choice. On the other hand, although there are many
large AS clusters in the big torrents, the locality of most peers is poor in nature.
herefore, the peers in a large AS cluster have both the potential and incentive to
=
i
N
/ +
i
q
