Transcript P2P Networks

P2P Networks
Advanced Computer Networks: Part 1
Agenda

What are P2P Systems?

File Sharing Techniques

Content Delivery

PageRank

Example of Work
2
What are P2P Systems?

Allow a group of computer users
(employing the same networking software)
to connect with each other to share
resources.

Processing power, disk storage, network
bandwidth and files to be directly available
to other peers.

They behave in a distributed manner
without a central server.

Peers can act as both server and client
then they are also called servent, which is
different from the traditional client-server
model.
3
What are P2P Systems?

First, protocols are constructed at
the application layer of the network
protocol stack.

Second, in most designs peers have
a unique identifier, which is the peer
ID or peer address.

Third, many of the message types
defined in various P2P protocols are
similar.

Finally, the protocol supports some
type of message-routing capability.

That is, a message intended for one
peer can be transmitted via
intermediate peers to reach the
destination peer.
4
What are P2P Systems?

An overlay network design that is consistent
with self-organization would not use a star
topology or a broadcast topology to operate
the peers or form the overlay.
 Instead the topology is likely to be
decentralized such that the
interconnectedness of any peer, referred to
as the degree, doesn’t dominate the overlay
graph.

Self-organization means that peers
cooperate in the formation and maintenance
of the overlay, with each peer using local
state and partial information about the
overlay.
5
What are P2P Systems?

Each peer determines its capabilities
based on its own resources.

Each peer also determines when it joins the overlay,
what requests it makes to the overlay, and when it
leaves the overlay.

A peer that searches for an object and
doesn’t find it might not be able to
determine whether the object doesn’t exist
in the overlay or the peer storing the object
has left the overlay.

Peers may act to limit their resource
contribution to the overlay, for example, by
disconnecting from the overlay when not
using it.
6
P2P File-Sharing Application

Peer-to-Peer technology is a way to share
music, video and documents, play games,
and facilitate online telephone conversations.


The technology enables computers using the same or
compatible P2P programs to form a network and share
digital files directly with other computers on the network.
User have digital media files that they want to
share with others.

These files are registered by the user using the local
application according to properties such as title, artist, date,
and format
7
P2P File-Sharing Application
How are P2Ps Different From
Traditional File Transfer Methods?

File transfer using the FTP is
referred as Client-Server file
transfer.

The file is downloaded from a
single source (the file server) by
the destination/PC (a client).
http://propakistani.pk/2010/02/08/what-is-peer-to-peer-p2p-filesharing-and-how-does-it-work/
8
P2P File-Sharing Application
How to Search for Files?

Users anywhere on the Internet can search
for these files by providing a query in terms of
some combination of the same attributes.


Peer that has local media files matching the
query will return information on how to
retrieve the files.


Query is sent to other online peers in the
network.
It may also forward the query to other peers.
Users may receive multiple successful
responses to their query and can then select
the files they want to retrieve.

The files are then downloaded from the remote
peer to the local machine.
9
P2P File-Sharing Application
Challenging Issues

A large amount of polluted or corrupted content
has been published in file-sharing systems, and it
is difficult for a user to distinguish such content
from the original digital content they seek.

A free rider is a peer that uses the file-sharing
application to access content from others but does
not contribute content to the same degree to the
community of peers.
10
P2P File-Sharing Application
Proxy Connection
 If the user hasn’t initiated any file searches or
selected local files for sharing, the P2P application is
most likely using the computer and its network
connection for other peers.



It may be responding to search requests from other peers
or acting as a bootstrap peer for newly joining peers.
It may be caching popular files on its disk drive to save
search time for other peers.
It could be proxying connections on behalf of peers
that are behind firewalls.
11
P2P File-Sharing Application

Peer is connected to the overlay, it can exchange
messages with other peers in its neighbor list.


Over time it can add and remove peers from its neighbor list.
To refresh and update its neighbor list, it can send requests to
current neighbors asking for their neighbors.
12
P2P File-Sharing Application

Random graphs and scale-free or power-law random
graphs are relevant to most unstructured topologies.
13
File Sharing Techniques (Gnutella)

Gnutella was the first full P2P file-sharing system
and has remained one of the more popular systems
to date.

The earliest versions of the Gnutella protocol,
through version 0.4, used an unstructured overlay
with flooding for query routing.

After scalability became an apparent performance
issue, the most recent version of the Gnutella
protocol (version 0.6) adopted a superpeer
architecture in which the high-capacity peers are
superpeers and all queries are routed, using a
flooding mechanism, between superpeers.

Gnutella network has a power-law distribution and
what the peer lifetime distribution is.
14
File Sharing Techniques (Napster)

Napster is a file-sharing system that used
a central server for storing and searching
the directory of files, but performed the file
transfer in a direct P2P fashion.

The majority of these designs used an
unstructured overlay mechanism.

Napster is not a full P2P system, it
popularized the P2P concept in the mass
media and influenced subsequent filesharing systems.
15
File Sharing Techniques (Freenet)

Peer identifiers are called routing keys.

Each peer has a fixed-size routing table that stores links to other peers.

Each entry contains the routing key of the peer.

Freenet uses key-based routing for inserting and retrieving objects in the mesh.

Requests are forwarded to peers with the closest matching routing key.

If a request along one hop fails, the peer will try the next closest routing key in its routing table.
16
File Sharing Techniques

Most existing P2P indexing schemes can be
categorized into local, centralized, distributed,
or hybrid indexing types.
17
File Sharing Techniques (Centralized
Indexing)

The index is kept on a centralized server.


Object lookup is done by searching over the index on the central server to
obtain the location of the target object.
Napster,190 the father of today’s P2P content distribution system, is a
classical example of centralized indexing.

Napster, peers share MP3 files stored locally on their hard drives.

Text-based content description, such as the title of a song, is then
generated, indexed, and stored by the Napster server.

Each peer in the Napster network uses the Napster client software to
connect to the centralized Napster server.

Peers connected to the Napster server can submit keyword-based queries
for a particular audio file.
18
File Sharing Techniques (Centralized
Indexing)

A list of matching files along with the description and
location of the file is sent to the peer from the
server.

The peer then tries to connect to the peer with the
desired audio file and transfers the target content in
a P2P fashion.

The most noticeable drawbacks of a fully centralized
indexing system include vulnerability to attacks on
the server and the possibility of a bottleneck at the
server.
19
File Sharing Techniques (Localized
Index)

Each peer keeps the directory of its own data objects locally.

The early Gnutella design is a typical example of local index-based P2P system.

Local object indexing can be used to support complex query search.



Compared to centralized indexing schemes, localized indexing can create more network
load since queries potentially have to be sent to many peers in the overlay.



When a peer generates a query, it conveys the query to peers in the network to locate the desired
object, most often through flooding or random walk.
Forwarding of query messages is stopped when the desired object is found or when the Time-to-Live
(TTL) value expires.
But since query processing is distributed across many peers in parallel, there is inherent scalability.
If any peer node is faulty, it affects the objects stored locally but not the overall search mechanism of
the network.
Localized indexing offers advantages over centralized schemes in terms of system
scalability and reliability. It has led to the development of many Gnutella-based systems
20
File Sharing Techniques (Distributed
Index)

Distributes the index over all peers.

Pointers to a single object may reside in multiple indices located in
different peer nodes, most often in neighborhood peer nodes.

Freenet is one of the earliest P2P systems in this category.

It uses content hash keys to identify files.

Queries are forwarded to neighborhood peers based on a peer routing
table until the target object is found or the TTL threshold is reached.

One disadvantage posed by Freenet is its long startup time at peer
joining.
21
File Sharing Techniques (Hybrid Index)

Hybrid indexing intends to take advantage of the
query efficiency of centralized indexing and the
scalability of localized and distributed indexing.

Superpeers (supernodes) maintain the indices in a
distributed manner.

A query from an ordinary peer is sent to its
superpeer to retrieve the location of the desired
object.

The superpeer forwards the query to other
superpeers if it does not find the desired object in
its own index.

Information about the desired object is sent back to
the ordinary peer via its superpeer.

Thereafter, the query issuer may directly contact
the peer with the desired object.
22
Content Delivery

The large and growing number of audiences and
traffic levels, content providers have tried to deploy
networks and systems that can provide increased
scalability, reliability, and quality of services.

A content delivery network or content distribution
network (CDN) is a system of computers containing
copies
of data
placed
at
various nodes of
a network.

When properly designed and implemented, a CDN
can improve access to the data it caches by
increasing access bandwidth and redundancy and
reducing access delay.

CDN is an overlay network that employs
technologies, such as caching, load balancing,
scheduling, and request routing, to push replicated
content close to the network

To reduce the cost of content delivery, today many
companies are seeking or are offering peer-to-peerbased solutions in both personal and enterprise
applications
23
Content Delivery (P2P Caching)

Caching is a well-known
strategy in many networked
applications.

The capability to:



Reduce origin server load and
bandwidth requirement
Reduce network bandwidth usage
Reduce client-side latency leads to
improved scalability and
performance with lower cost.
24
Content Delivery (P2P Caching)
25
PageRank
26
26
RFID Datenvolumen
Major Issues of Existing PageRank
 Only Structure Consideration
 Knowledge of Global Network Structure
 Broken Link Problems
27
27
Example of Work:
Extended Pagerank Calculation
28
28
Ideas: Structure and Network Parameters
Combination
 Pagerank can be calculated by random walkers
(visiting probability)
 Apply a population of random walkers to handle
the large network size
 Apply network parameters to modify the
transistion probability of random walks
29
29
Proposed Pagerank Calculation
 Transition probability of walkers
 Equal probability (without BW consideration)
 Non-equal probability (with BW consideration)
 Ranking calculation
30
30
Simulation Results (1)
PageRank
Proposed Work (without BW consideration)
31
31
Simulation Results (2)-without BW Consideration
One-Walker Convergence Time
Comparison in Different Size of
Network
Convergence Time
Comparison in Different Number
of Walkers
Where,
32
32
Simulation Results (3)-with BW Consideration
33
33
Heaps of Documents on Super Peers
34
34
Heaps of Documents
 Ant colony optimization: Dead ant collection
 Dropping probability
 Picking up probability
 Document clustering:
 Heaps of the same kind of documents
35
35
Heaps of Documents
t=1
t=10,000
t=500,000
36
t=100,000
t=811,042
36
Heaps of Documents
 Combine together between
0.01
0.008
0.006
0.004
0.002
0
20
15
20
15
10
10
5
5
0
0
To be robust search and file sharing system
37
37
Simulation Result
BW Structure
t=1
38
38
Simulation Result
t=10,0000
39
39
Simulation Result
t=100,0000
40
40
Simulation Result
t=500,0000
41
41
Simulation Result
t=503,663
42
42
Other Applications

VoIP (voice over IP) lack of voice
quality and limited availability of
broadband access in the consumer
market.

Skype is a VoP2P client

Skype offers both free desktop-todesktop calls and low-cost desktopto-public switched telephone
network (PSTN) calls, including
international calls.
43
Other Applications

P2PTV delivery often follows a channel organization in which
content is organized and accessed according to a directory of
programs and movies.
 Unlike file-sharing systems in which a media file is first
downloaded to the user’s computer and then played locally,
video-streaming applications must provide a real-time stream
transfer rate to each peer that equals the video playback rate.

If a media stream is encoded at 1.5 Mbps and there is a single
peer acting as the source for the stream, the path from the source
peer to the playback peer must provide a data transfer rate of 1.5
Mbps on average.

Some variation in the playback rate along the path can be
accommodated by pre-buffering a sufficient number of video frames.
Then if the transfer rate temporarily drops, the extra content in the
buffer is used to prevent dropouts at the rendering side.
44
Zattoo
Joost
45
Other Applications

Video-on-demand type applications, peers start
viewing a stream at arbitrary times, and those peers
that start viewing a stream concurrently may soon
diverge in stream position due to user actions such
as pause or rewind.

To avoid transferring complete copies of video files
to peers at playback, a method is needed for a peer
to locate the next chunk of video in its playback
schedule from some other peers in the P2P system.
46
Basic Security in P2P Networks
Advanced Computer Networks: Part 1
P2P Network

Network classification based on network
connectivity



Exponential Networks:
Homogenous network, [average] node
connectivity is equally distributed
Scale-free networks:
Follows power-law for connectivity, that is there
are some highly connected nodes and many not
to highly connected nodes
Current P2P systems are scale-free networks
Failure and Attack

Failure:


Attack:



Random failure of nodes and/or infrastructure elements
Systematic failure of nodes and/or infrastructure elements
Scale-free networks are failure-tolerance
Exponential networks are attack-tolerance
Why?
 Most P2P systems give priority for failure-tolerance
over attack-tolerance
Attack Classification

Infrastructure Attacks:

Attacks aimed at disabling p2p system
e.g: eliminating nodes, attacks on routing protocols

Semantic Attacks:

Attacks aimed at p2p system but not to disable
the system but to make users abandon the
system
e.g: bad content, asymmetric consumption

Both attacks are equally effective because
p2p is a “peoples’ system”
Attacks & Defenses

Attack detection & recovery involves…





Infrastructure Attacks



Identifying Invariants in the System
Monitoring the Invariants
Detecting/Ascertain Attacks
Triggering Recovery Procedure
Attacks on nodes
Attacks on routing mechanism
Semantic Attacks



Storage & Retrieval Attacks
Flooding
Face/Off
References

John Buford, Heather Yu, Eng Keong Lua , P2P
Networking and Applications, Morgan Kaufmann Series
in Networking, December 26, 2008.

Kulesh Shanmugasundaram, Peer-to-Peer
Systems:Security Issues, presentation (ppt).
52