what-when-how
In Depth Tutorials and Information
Step.3
: If this value is smaller than a preconfigured threshold
e
, send the peer set
information (the sets of neighbor peers) to peer
x
by giving high priority to
the neighbors that are in the same AS with
x
; or else send the peer set infor-
mation using random strategy.
We processed a simple validation of this proposed method by few AS information
within three torrents. Both inputs and outputs are shown in Table 6.4, where the
upper table is the table
T
and the lower table shows the entropy value of different
ASes. According to the outputs, we learn that peers in AS# 3352, 2119 and 6461
are more likely to form a large AS cluster than that of others.
his is obviously not a perfect result because the data in table
T
is limited.
However, it still seems somewhat reasonable because Table I can validate some of
its results, where the AS# 3352, 2119, and 6461 are all very popular ASes. Note that
although As popularity can provide some meaningful information for the valida-
tion, it not feasible for the peer prediction. Obviously, the variation of peer number
cannot reflect the relationship between ASes; In particular, AS# 3662 and AS#
6461 have very similar popularity in Table 6.1; yet the peers inside these ASes are
not necessarily having similar clustering properties.
6.4.3 Summary
In this chapter, we have discussed the interactions of BitTorrent clients, from
microscope piece level to intra-torrent and inter-torrent levels. We have presented
measurements on the piece populations in torrents, and investigated the effective-
ness of the rarest-first policy for piece replication from a piece distribution and
evolutionary perspective. We have shown that the policy is quite effective once all
pieces become available in the system, and throughout the lifetime of the torrent.
However, some deviations from the ideal were apparent soon after creation of the
torrent, and in some of the larger torrents studied.
We have also shown experimentally the network evolutions in a torrent, which
we found change significantly over time. We have quantified the scale-free nature of
one of the networks in a torrent, that of peers unchoking each other. We found that
this scale-free nature is independent of time and the changing parameters of the
experiment. We have also gone beyond previous studies to show that, after the very
early stages, most of the networks in a torrent are purely random graphs with no
clustering present. herefore, the graphs do not exhibit the small-world character-
istics that have been found in many other peer-to-peer network overlays. However,
the ever-present churn in a BitTorrent system may have impacted this result.
We have therefore successfully designed a tracker modification to introduce
small-world networks into a torrent. he modiication has been tested, both through
simulation and experimentally, to have introduced a large amount of clustering, at
the expense of only a small increase in the characteristic path length. Our changes
