ECE6160: Advanced Computer Networks PeertoPeer P2P Networks

1
ECE6160 Advanced Computer Networks
Peer-to-Peer (P2P) Networks

• Student Dina Machuve
• Instructor Dr. Xubin He
• Term Fall 2007

2
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

3
Overview of P2P Networks

• Peer-to-Peer Networks consist of multiple clients
  forming a network where peer nodes function as
  both clients and servers to other nodes in
  the network
• Used for connecting nodes via largely ad hoc
  connections
• All clients provide resources that include
  bandwidth, storage and computing power
• Highly scalable. Total capacity of system
  increases with increasing demand
• Robust in case of network failure

4
P2P Classes

• Pure P2P System
• Peers share data without a centralized
  coordination
• Hybrid P2P system
• Some operations are intentionally centralized,
  such as indexing of peers files.

5
Network Application Architectures

6
P2P Network Applications

• P2P File-Sharing
• Media Streaming (audio and video)
• Instant messaging
• Voice over IP (VOIP)

7
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

8
P2P Network File-Sharing

• P2P file-sharing is a popular Internet
  application
• Sharing of MP3 files (3-8Mb), Videos (10-1000Mb),
  software, documents and images.
• Accounts for a major fraction of all internet
  traffic (in 2004)
• File-sharing applications include Napster, KaZaA,
  eDonkey and Gnutella.
• BitTorrent is a P2P protocol essentially used for
  downloading files, it doesnt provide search
  mechanism.

9
Content Location Methods

• A large number of peers are connected in a P2P
  file-sharing system and continuously disconnect
• content location is essential to determine the IP
  addresses of connected peers that have copies of
  desired object.
• Three architectures for content location
• Centralized Directory
• Query Flooding
• Exploiting heterogeneity

10
Centralized Directory

• Used by the defunct Napster
• Centralized directory server keeps an updated
  database of IP address, names of objects
  available for sharing of connected peers
• The file transfer between peers is decentralized
• The process of locating content is highly
  centralized

11
The P2P paradigm with a centralized directory
12
Query Flooding

• Used by Gnutella, public domain file-sharing
  application
• Uses a fully distributed architecture
• File-sharing and content search is decentralized
• Unstructured P2P system where all peers are
  involved in query flooding
• Peers forward query messages to each other on the
  overlay network, referred to as query flooding.
• When a requesting peer receives query hit
  messages, a direct TCP connection is established
  and file-sharing begins

13
Content search and file transfer in Gnutella
14
Exploiting Heterogeneity

• Used by KaZaA
• In April 2004, KaZaA had more than 3million users
  sharing over 5,000 terabytes of content
• A small fraction of the more powerful peers are
  designated as group leaders
• Higher bandwidth and Internet connections and
  have greater responsibilities
• Group leader has up to a few hundred children
  peers
• Peers form direct TCP connections with one of the
  group leaders that serves the peers

15
Hierarchical overlay network for P2P file sharing
16
Message Duplications

• The mechanism of a peer randomly joining and
  leaving a P2P network has negative impact
• Mismatch problem between a P2P overlay network
  and the underlying physical network topology
• Query flooding causes unnecessary traffic
• Only 2-5 of Gnutella traffic link peers, and
  more than 40 of all Gnutella peers are located
  within the top 10ASs.
• Same message traverse the same physical link
  multiple times causing a large amount of
  unnecessary traffic

17
Topology Mismatch Problem Scenario

• Mismatching overlay
• Matching overlay
• Underlying physical topology

18
Adaptive Connection Establishment (ACE)

• ACE is an algorithm proposed by L. Xiao et al. to
  minimize the effect due to topology mismatch
• Builds an overlay multicast tree among each
  source node and the peers within a certain
  diameter from the source peer
• Involves three phases, namely neighbor cost table
  construction and exchanging, selective flooding
  and overlay optimization
• Optimizes the neighbor connections that are not
  on the tree while retaining the search scope
• The approach effectively solves the mismatch
  problem and reduce P2P traffic noted by reduced
  cost of query reaching peer nodes (65) and
  reduced query response time (35).

19
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

20
Peer-to-Peer Media Streaming

• Media Streaming applications
• Internet Television (IPTV)
• Video conferencing
• Video streaming
• Video-on-demand
• Audio streaming
• Voice and Video over IP - Skype

21
Motivation of Media streaming

• Growth of the Internet and multimedia
  communications
• Media streaming uses P2P Technology
• Bandwidth scalability
• Cost-effective networks
• Distributed storage
• Increased computing resources

22
Large scale P2P media streaming schemes

• PPLive
• Very popular and free P2P-based IPTV application
• Provides service to over 400,000 users daily
• GridMedia
• Large-scale deployment of P2P live-video
  streaming application
• Streaming achieved at a constant rate of 300-500
  kbps and 600 kbps for high definition 2
• Adopted by CCTV in 2005 and 2006, broadcasted
  live Chinese Spring festival show supporting more
  than 500,000 and 1,800,000 respectively 2

23
Media streaming schemes

• SopCast
• IPTV application that supported more than 100,000
  simultaneous users after its initial deployment
• Provides services to consumers and service
  providers
• Coolstreaming
• The earliest large-scale P2P live-streaming
  system
• Developed in 2004, concurrent users reach over
  80,000 with an average rate of 400 kb/s
• On the consumer domain
• Mobile communicators, and networked consumer
  electronics

24
Normalized search volume for popular P2P schemes

• source www.google.com/trends

25
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

26
Real-Time Streaming Protocol (RTSP)

• RTSP is an application-level protocol for control
  over the delivery of data with real-time
  properties (RFC 2326) such as audio and video.
• allows a media player to control the transmission
  of a media stream
• RTSP messages are sent over either TCP or UDP.
• RTSP Control commands sent by client to the
  server include SETUP, PLAY, PAUSE and TEARDOWN.

27
Video streaming server and client 1, 3

• The server will use the RTP to packetize the
  video for transport over UDP
• Client is setup to use RTSP to control actions on
  the server

28
RTSP Video streaming

• In playing state
• Server periodically grabs a stored JPEG frame,
  packetizes the frame with RTP, and sends the RTP
  packet into a UDP socket
• The client receives the RTP packets, removes the
  JPEG frames, decompresses the frames, and renders
  the frames on the clients monitor

29
Interaction between Server and Client
30
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

31
IP Multicast

• The earlier P2P streaming applications were
  largely built on IP multicast framework
• Extension of the traditional best-effort unicast
  model to multi-point packet transmissions
• Dissemination of packets to destinations achieved
  through the construction of a spanning tree
  across routers in networks
• IP Multicast operates at the network level of the
  IP Stack

32
Drawbacks of IP Multicast

• Scalability in that there are potentially a large
  number of multicast groups that must be managed
  in a large network
• Complexity in coordination of dynamic spanning
  tree (s) at routers across different autonomous
  subnets
• Routers must maintain the state, which violates
  the stateless principles and creates difficulty
  in the design of high-level functions such as
  error, flow and congestion control

33
Overlay network

• Application-level or end-system multicast
  addressed the classical IP multicast problems
• Used overlay network, a virtual topology over the
  unicast Internet
• Greater flexibility, all the nodes have strong
  buffering capabilities and can adaptively
  determine the data forwarding directions

34
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

35
Data-Driven Overlay Network- DONet

• A data-centric design of a streaming overlay, in
  which every peer node periodically exchanges data
  availability information with a set of peers, and
  retrieves unavailable data from one or more
  peers, or supplies available data to peers.
• It eliminates the requirement for constructing
  and maintaining any specific overlay network.
• A public Internet-based DONet implementation is
  Coolstreaming v0.9

36
Advantages of DONet

• Easy to deploy
• There is no need to maintain any global structure
• Efficient
• Data forwarding is not determined prior to, but
  instead on, availability
• Robust and resilient
• The peer nodes and data availability information
  are dynamically and periodically updated, so the
  system can be self-evolving

37
DONet Node Architecture

• Membership Manager
• Helps the node maintain a partial view of other
  overlay nodes
• It consists of a membership cache (mCache) that
  has a partial list of the identifiers for the
  current active nodes in the system
• Partnership Manager
• Establishes and maintains the partnership with
  other nodes
• Buffer Map (BM) represents availability of
  segments in the buffer and BM are exchanged among
  the peers periodically
• Scheduler
• Schedules the transmission of video data
  segments.

38
Node Joining Algorithm

• A new node first establishes connection to the
  origin node (A)
• The origin node randomly selects another node
  (deputy node) from its mCache and redirects the
  new node to the deputy node
• The new peer node obtains a list of peers from
  the deputy node and contacts these peers and
  establishes its own peers on the overlay
• The mCache is then updated using membership
  messages

39
Partnership issues in DONet

• The initial node joining process takes an
  excessive amount of time (10-20seconds)
• The random peer selection causes a new node to
  connect multiple times before it can successfully
  establish stable connection

40
Live Media Streaming over DONet nodes

• A video stream is divided into segments of
  uniform length
• The availability of the segments in the buffer is
  represented by a Buffer Map (BM)
• Each node exchanges its BM with the peers
  continuously, and then a schedule is generated
  for fetching segments from partners accordingly

41
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

42
Architecture module of DONet experiment system

• DONet system
• Implemented in Python programming language
• Concurrent operations are implemented using event
  queue with non-blocking sockets
• Console and Automaton module
• Console is the interactive command interface for
  the whole system
• Automaton in the console is used to launch
  experiments and execute commands predefined in a
  queue
• Command Dispatcher and Report Collector
• Each command message has unique sequence number

43
Experiment Setup

• The origin node is located in PlanetLab, USA. The
  monitoring node is located in Hong Kong. Access
  to PlanetLab is achieved through remote logins
• Experiments involve a total number of PlanetLab
  active nodes ranging from 200 to 300
• The distributed testbed is highly scalable and
  extensible to add new nodes or new features due
  to the design of module architecture, tools
  provided by PlanetLab.
• Under stable environment, all nodes join in an
  initialization period (around 1min) and then
  persist for 120min (streaming lifetime)
• Default streaming rate is 500Kbps. Each segment
  contains 1 sec of the stream and DONet node
  maintains a sliding window of 60 segments.

44
Control Overhead

• Control overhead Control traffic volume/Video
  traffic volume at each node
• Control messages in DONet are for exchanging data
  availability information. Hence the number of
  partners becomes a key factor to the control
• The overhead increases with an increase of the
  number of partners.
• Control overhead at each node is almost
  independent of the overlay size because BM are
  only locally exchanged.

45
Playback Continuity

• Continuous playback is a primary objective for
  streaming applications.
• Continuity index is the measure for continuity
  evaluated as the number of segments that arrive
  before or on playback deadlines over the total
  number segments.
• The continuity improves with increasing number of
  partners, since each node has more number of
  suppliers.

46
Playback Continuity

• Continuity index as a function of different
  streaming rates for an overlay size of 200 nodes.
• DONet is scalable in terms of both overlay size
  and streaming rate
• Practical number of partners is 4

47
Performance under Dynamic Environment

• DONet with dynamic node joining, leaving and
  failing
• ON period node actively participates the overlay
• OFF period node leaves (or fails) the overlay
• DONet remains acceptable even under highly
  dynamic networks
• Control overhead is slightly higher with a
  shorter ON/OFF period (more dynamic node
  behavior)
• A shorter ON/OFF period leads to poor continuity

48
Comparison with Tree-based overlay

• The degree of both DONet node and tree-based
  overlay used is 4.
• The end-to-end delays for delivering each segment
  are measured using overlay hop-count
• Under both stable and dynamic environments, the
  delay measures of the tree-based overlay are
  higher than DONet.
• This is due to out-bound bandwidth constraints
  that increase the height of the tree

49
DONet vs. Tree-based Overlay

• The continuity index of the tree topology is very
  low compared to DONet
• Tree topology is vulnerable to internal node
  failures

50
Playback discontinuity

• The tree is full and balanced
• Playback discontinuity is caused by node
  departures and failures
• DONet achieves much better playback continuity

51
Outline

• Introduction
• P2P File-Sharing
• P2P Media Streaming
• Real-Time Streaming Protocol
• IP Multicast
• DONet
• Analysis of DONet P2P scheme
• Conclusion

52
Conclusions

• DONets control overhead is reasonably low,
  around 1 of the video traffic and this remains
  unchanged with an increase of the overlay size.
• This makes DONet acceptable for live media
  streaming.
• DONet system is scalable. The automatic command
  dispatching and report collecting system allows
  for launching the DONet program to more nodes.
• DONet delivers better playback quality as
  compared with tree-based overlay in terms of more
  continuous streaming with comparable delay.
• The PlanetLab has offered stable network
  condition that the reproducibility problems were
  not very severe.

53
References

• J. Kurose and K. Ross, Computer Networking A
  Top-Down Approach Featuring the Internet 3rd
  Edition, Pearson Education Inc., 2005
• Y. Tang, J.G. Luo, Q. Zhang, M. Zhang and S.Q.
  Yang, Deploying P2P Networks for Large-Scale
  Live Video-Streaming Service, IEEE
  Communications Magazine, Vol. 45, No.6, June
  2007, pp. 100-106
• http//oregonstate.edu/lemhachr/classes/multimedi
  a-networking-streaming.phpfeatures
• X.Zhang, J. Liu, B. Li and T.P. Yum,
  CoolStreaming/DONet A Data-Driven Overlay
  Network for Efficient Live Media Streaming, IEEE
  INFOCOM, 2005
• B. Li and H. Yin, Peer-to-Peer Live Video
  Streaming on the Internet Issues, Existing
  Approaches and Challenges, IEEE Communications
  Magazine, Vol. 45, No.6, June 2007, pp. 94-99
• L. Xiao, Y. Liu, and L.M. Ni, Improving
  Unstructured Peer-to-Peer Systems by Adaptive
  Connection Establishment, IEEE Transactions on
  Computers, Vol.54, No.9, 2005, pp.1091-1103.

54
Thank You