Information Technology Reference
In-Depth Information
P2P brings significant challenges in trac engineering for ISP (Internet Service
Provider)[3].
Because P2P applications are ignorant of the underlying network topology,
most P2P applications apply application-level routing that is based only on the
overlay network metrics [5]. Moreover, some P2P applications select the source
of their downloading randomly, which may lead a P2P user in New York City
to download from a user in Los Angeles, while this kind of data is available in
New York City or in Washington DC. This kind of long-distance downloading
may decrease the network eciency and the performance of the P2P applications
concurrently. P2P applications can avoid this by selecting neighbors with a lower
delay or less router hops, but purely selecting neighbors with a low delay or less
router hops may cause a Comcast user to select an AT&T user as its neighbor.
This cross-ISP neighboring will generate unnecessary interdomain trac, thereby
significantly increasing the operational costs of ISPs. In conclusion, current P2P
applications have the following problems:
(1) A P2P system may cause the dispersion of network trac and make the
trac to unnecessarily flow through multiple intra-domain links. By conducting
practical tests, [6] found that every bit of P2P trac in Verizon needs 5.5 hops
when passing through 1000 miles on average, and this average number of hops
can be reduced to 0.89 without compromising the P2P application's performance.
(2) A P2P system may generate massive inter-domain trac or cause massive
trac that when produced by multiple ISPs pass through a specific network [9].
In [8], Karagiannis studied the BitTorrent performance in a college network. He
found that this low-eciency inter-domain trac may cause significant financial
losses for ISPs. Even in the case of the top-level ISP (tier-1 ISPs who do not
pay other ISPs), the inter-domain tra c caused by a P2P system can cause
the tra c between these tier-1 ISPs to lose balance and then violate the P2P
protocol.
However, a one-sided strategy of P2P and ISP is not ideal. As a result, a
cooperative model of P2P and ISP should be built, making both sides exchange
information and control the P2P trac cooperatively, and thereby improving
the network eciency and the P2P performance simultaneously.
1.2 Related Works
[4] studied the advantages and disadvantages of P2P, and proposed a middle
server called ” oracle ”, by which ISPs can provide a neighbor selection policy
for P2P users. After a peer sends its list of potential neighbors to ” oracle ”,
oracle ” will sort all the possible neighbors according to certain criteria, such as
the nearest principle and the link bandwidth. The sorted neighbor list will guide
peers to select neighbors and improve the P2P performance. At the same time,
the ISPs can effectively manage massive the P2P trac with this mechanism,
assuring that the trac does not pass across them and that it is led to the right
path. With the abovementioned mechanism, ISPs can provide a better network
service for their users.
 
Search WWH ::




Custom Search