Warren Kenny,

1
Peer-to-Peer in MANETs The State of the Art

• Warren Kenny,
• Distributed Systems Group,
• Trinity College, Dublin
• 13.03.2008

2
Presentation Format
3
Peer-to-Peer Protocol Types

• Unstructured
• Links are formed arbitrarily, with no relation
  between the content held by nodes and their
  address in the overlay
• Content retrival is generally accomplished
  through query flooding
• Structured
• Uses a globally consistent protocol to route
  content requests to appropriate nodes
• Nodes often have identities with no relation to
  their physical position in the network, but
  rather to the content they hold

4
Unstructured Overlay Protocols

• Gnutella v.1 1
• Developed to facilitate decentralized
  file-sharing on the internet
• Nodes bootstrap on startup and find at least one
  other node through various means ( cache,
  pre-designated address lists, even IRC )?
• Nodes will continue the connection process until
  they have accrued a specified number of
  neighbours
• Queries are flooded through a node's neighbours,
  who then query their own neighbours. Queries are
  TTL-limited
• Scales poorly as more peers join the network,
  query floods consume more and more time and
  bandwidth
• Now obsolete, succeeded by Gnutella v.2

5
Unstructured Overlay Protocols

• Gnutella v.2 2
• Improves upon Gnutella v.1 primarily in terms of
  scalability and query speed
• Nodes with sufficient connection quality and
  power are designated Ultra Nodes
• Promotion occurs on-demand or voluntarily
• Standard client nodes are referred to as Leaf
  Nodes
• Leaf Nodes send queries to the Ultra Node to
  which they are connected
• Ultra Nodes forward queries to other Ultra Nodes
  with a low TTL
• Leaf Nodes inform their Ultra Node parent exactly
  which files they hold, allowing Ultra Nodes to
  request files only from those leaves which hold
  them
• Scales far better than Gnutella v.1

6
Unstructured Overlay Protocols

• Bit-Torrent 3
• Peer-to-Peer File Sharing Protocol,
  Semi-Decentralized
• Peers find files through external means ( web
  search, etc. ) and download a .torrent file which
  details the connection information for a swarm
• Swarms are composed of trackers, leechers and
  seeders
• Trackers act as coordinators for intra-swarm
  connections and peer joins, as well as moderators
  which ensure fair bartering
• Leechers request and download file chunks from
  other leechers and seeders
• Once a leecher has retrieved the entire file, it
  becomes a seeder and uploads selflessly to other
  peers until it is disconnected
• Encourages fair network use through a bartering
  system
• Tracker settings determine the UD ratio required
  for continued use

7
Unstructured Overlay Protocols

• CYCLON 4
• Gossip-based unstructured overlay protocol
  designed to reduce member management overheads
• Uses a variation on the Basic Shuffling algorithm
• Each node is aware of a small number of its
  neighbours
• Periodically, neighbours will communicate in
  order to swap relationships, passing neighbours
  among themselves and shuffling the overlay
  topology
• Enhanced Shuffling is a modification of this
  protocol which results in neighbours only
  exchanging peers with their oldest neighbour
• Limits the time a node can be passed around until
  it is chosen to take part in an exchange,
  resulting in greater fairness
• Results in an overlay which can cope easily with
  node addition and removal while requiring little
  overhead

8
Unstructured Overlay Protocols

• T-Man 5
• Uses a ranking function to allow nodes to order
  their neighbours according to a target topology
• Designed to make overlay restructuring simple and
  efficient
• Nodes maintain views composed of neighbour node
  addresses and profiles ( geographical location,
  ID, etc. )?
• Nodes exchange views with their neighbours so
  that they can better evaluate alternative
  connections based on the target topology
• Allows for the construction of topologies
  representing Lines, Tori, Binary Trees, etc.
• Scales well and copes with node addition/removal

9
Structured Overlay Protocols

• Pastry 10
• Each node in the network has its own unique Node
  ID a 128-bit key which indicates its position in
  a circular ID space
• Nodes which receive a key-message pair can
  normally route that message to the node with the
  nearest key in O(logN) steps where N is the
  number of Pastry nodes in the network
• Each node maintains a routing table, a
  neighbourhood set and a leaf set
• The routing table contains the Node Ids of other
  nodes organised in terms of prefix equality
• The neighbourhood set contains the Ids of the
  nodes in closest physical proximity
• The leaf set contains the Node Ids closest
  numerically to the host's
• Messages are routed using the leaf set and
  routing table

10
Structured Overlay Protocols

• Chord 11
• Similar to Pastry, but uses a simpler table
  system and handles joins and departures more
  efficiently
• Chord provides more consistent hashing than
  Pastry, constructing a hash using the SHA-1 of
  the node IP
• The key-value pair corresponding to some content
  is assigned to the node with same or greater ID
• Chord uses either a simple ring-based search
  algorithm or a scalable key location algorithm
  which uses a lookup or finger table to assist
  routing
• Chord copes with joins and departures very
  simply each node periodically updates its
  successor ID and finger tables
• Each node also keeps a list of its first r
  successors so that it can quickly switch to a new
  successor if it loses contact

11
Peer-to-Peer Applications

• Distributed Web Cache ( Structured )?
• Squirrel
• Distributed File System ( Structured )?
• Ivy
• Peer-to-Peer File Sharing ( Unstructured )?
• Gnutella
• Peer-to-Peer Telephony ( Unstructured )?
• Skype
• Distributed Information Retrieval ( Structured )?
• pSearch

12
Squirrel Distributed Web Cache 6

• Installed on client machines and used as
  intermediate proxy
• Object-IDs formed by hashing request URLs
• Uses DHTs to map URL/Object-IDs to specific peers
  and make requests
• Sends a request for the cached content to the
  peer with the nearest ID to the object ID
• Scales up to thousands of nodes in simulation
• Nodes joining the network are given the objects
  they must store by nodes which previously held
  them
• Nodes leaving the network have their objects
  handled by nodes with the next-nearest ID

13
Ivy Distributed Read/Write FS 7

• Nodes store local logs which detail file system
  modifications
• Stored in a DHash DHT
• Nodes find data by reading all logs but append
  data by writing only to their own log, allowing
  for consistency without locking
• Resolves conflicts using a complex versioning
  system and application-specific resolvers
• 2-3 times slower than NFS due to network and
  hashing overheads

14
Gnutella P2P File Sharing Network 2

• Origin of the Gnutella Protocol
• A true decentralized network, very difficult to
  shut down due to the lack of a centralized
  essential authority
• Composed of leaf nodes and ultra nodes
• Optimizes querying by limiting the number of
  results returned to the leaf node
• Accomplished by imposing a low TTL on queries
• Ultra Nodes compare the number of returned
  results to the number of nodes queried and
  designate the query as rare if the ratio between
  results and queried nodes is low
• Rare queries are forwarded to more Ultra Nodes in
  order to get more results

15
Skype P2P Internet Telephony 8

• Composed of 3 node types
• Standard Node A computer with a Skype Client
• Super Node A hub host which facilitates
  connections between standard nodes , any client
  with sufficient connection quality and power can
  be one
• Login Node Controlled by Skype, handles logins
  and registrations, ensuring user uniqueness and
  identity
• Standard Nodes store Super-Node IP addresses in
  their Host Cache a list of super-nodes which is
  refreshed regularly
• Super-Nodes are the targets for queries by
  Standard Nodes
• Super-Nodes cache query results
• Super-Nodes are organized into slots of 9-10 and
  blocks of 8 slots to facilitate organised querying

16
pSearch P2P Information Retrieval 9

• Creates a semantic overlay in a node network
• Contents are organized around their semantics (
  e.g distance in terms of routing hops
  proportional to the difference between documents
  )?
• This turns the network into a Content-Addressable
  Network or CAN
• Provides a DHT abstraction over Cartesian space
• Involves partitioning the network into zones,
  each of which is controlled by a node and
  identified by a certain key range
• Semantics are generated using Latent Semantic
  Indexing
• New documents are analyzed using LSI and stored
  at the node with the nearest semantic index
• Queries are also assessed using LSI and are
  routed to the node with the nearest semantic
  index before being flooded to other nearby nodes.
  Results are returned to the origin

17
MANETs Protocol Types

• Reactive
• Routes are discovered on-demand
• Flow-Oriented
• Packets are routed along existing paths or
  on-demand otherwise
• Pro-Active
• Nodes flood the network with routing data
  continuously
• Hybrid
• Combination Reactive/Pro-Active protocols
• Geographical
• Takes into account the geographical location of
  nodes when determining routing paths

18
Reactive Routing Protocol

• Ad-Hoc On-Demand Distance Vector Routing 12
• Designed to reduce network overheads produced by
  pro-active routing
• When a node broadcasts a route request,
  neighbouring nodes record the request and
  re-broadcast it
• Results in an explosion of route requests
• When the request encounters a node which has a
  route to the destination node, confirmation is
  sent back to the requester
• The requester will choose the route which
  involves the fewest hops
• If a hop in the route fails, the process begins
  again
• Sequence numbers are used to prevent request
  re-broadcasts
• Unlike DSR, only the destination and sequence
  number information is stored in the route
  request/reply packet, reducing overheads
• Route discovery is slower than in pro-active
  protocols, but requires lover overhead

19
Flow Oriented Routing Protocol

• Multi-Path On-Demand Routing 13
• Designed primarily for sensor networks with a
  view to increasing energy efficiency
• Similar to AODV, MOR nodes flood the network when
  they require a route to another node
• In a sensor network, we are primarily concerned
  about routing data to and from a base station
• In a MOR-based network, the base station
  periodically floods the network with route
  requests
• Messages to and from the base station are
  recorded by intermediates and used to form routes
• Uses Unicast wherever possible instead of
  flooding
• Nodes record as many equal-hop routes as possible
• Results in lower energy consumption

20
Pro-Active Routing Protocol

• Optimized Link-State Routing 14
• Each node in the network periodically exchanges
  link-state messages with its neighbours
• Minimizes overheads by employing Multi-Point
  Replaying (MPR), under this scheme, each node
  chooses a sub-set of its neighbours as MPRs
  nodes which will re-transmit its link state
  messages
• Nodes which receive link-states from the origin
  node but which aren't MPRs for that node will
  record the data but won't retransmit
• Using exchanged link-state messages, nodes
  calculate the best routes to take in order to
  send data to each other node within the network
• Packets can be sent to their destinations with
  little or no delay
• No requirement for on-demand flooding
• Network overheads are higher than in a network
  utilizing a pro-active routing protocol, but
  transmission delays are lower

21
Hybrid Routing Protocol

• Zone Routing Protocol 14
• Combines pro-active and reactive route discovery
• Each node has a routing zone ( defined by a
  maximum hop count ) which defines the area within
  which they are required to maintain links
  pro-actively
• Nodes outside of this routing zone can only be
  reached through on-demand flooding
• Reduces overheads since the request is only
  broadcast from nodes at the boundaries of the
  routing zone
• Reduces overheads associated with pro-active
  protocols since link-state is only maintained
  within a limited range
• Depending on the size of the network, the
  protocol may become completely pro-active ( small
  networks ) or reactive ( large networks )?

22
Geographical Routing Protocol

• Greedy Perimeter Stateless Routing 15
• Protocol designed for networks where information
  regarding node geographical relationships is
  available
• When constructing routes, nodes exercise greedy
  forwarding
• Nodes make locally optimal decisions regarding
  where to forward packets
• Nodes will attempt to forward packets to the
  neighbour which is closest geographically to the
  destination
• Nodes need only hold topological information
  regarding their immediate neighbours in order for
  this protocol to work
• When no closest neighbour exists, the system
  enters perimeter forwarding mode in order to
  defeat the local maximum problem
• Nodes forward packets to the first node
  counter-clockwise to the imaginary line between
  the source and destination. This process is
  repeated by intermediate nodes using the
  right-hand rule
• When the packets enter a region in the same
  geographical location as specified in their
  headers, they are then returned to greedy mode

23
Mobile Ad-Hoc Network Applications

• Hybrid Networks 16
• Allow mobile users to stay in contact with
  infrastructure networks, despite changes in
  position
• Sensor Networks 17
• Monitor interesting phenomena by deploying large
  numbers of sensors which stay in contact via
  wireless connections
• Vehicular Networking 19
• Enable inter-vehicle communication to report
  traffic conditions, accidents, etc.
• Multi-Player Gaming 18
• Allow players using mobile gaming devices to play
  together without the need of intermediate
  infrastructure
• Emergency and Battlefield Communications 16
• Allow soldiers and emergency workers to share
  data about their surroundings and situation

24
Peer-to-Peer in Mobile Ad-Hoc Networks

• PDI General Purpose Lookup System 21
• Epidemic Dissemination of Key-Value pairings
• Uses agressive caching in order to reduce
  overheads associated with query flooding
• Popular query results are spread throughout the
  network in (k, v) pairs and cached by receiving
  nodes
• Query results are broadcast from their source
  node and cached by nearby nodes
• Invalidates stale cache entries both pro-actively
  and reactively
• Each cache entry has an associated time-out value
  determined by the source node
• This time-out is reset upon contact with the
  source node
• When the time-out reaches 0, the cache entry is
  deleted
• Source nodes can also issue cache entry Death
  Certificates
• Death certificates are associated with (k, v)
  pairs, cached by neighbours and sent to any node
  which requests the same (k, v) pair
• This prevents data resurrection a common
  occurrence in E.D-overlays

25
Peer-to-Peer in Mobile Ad-Hoc Networks

• ORION P2P File Sharing over MANETs 22
• Queries are flooded through the network
• Each query has an associated source ID and
  sequence number in order to prevent echoing
• Responses are recorded by intermediate nodes for
  future reference
• Allows intermediate nodes in a query cycle to
  record the same information as the query source,
  gaining information for free
• Nodes keep a cache of first-hop response sources
  on a per-file basis
• Reduces overheads associated with recording full
  routes within response packets
• Nodes transfer files in discreet blocks on a
  per-request basis
• Nodes compensate for route failure by re-routing
  requests to nodes which hold the same file

26
Peer-to-Peer in Mobile Ad-Hoc Networks

• 7DS Data Exchange in Disconnected Networks
• Allows nodes to retrieve data even when
  disconnected from a gateway
• Proxy and multicast search engine which attempts
  to find requested data on neighbouring nodes
  before consulting a gateway
• All nodes use a lightweight web server to act as
  providers for requested pages which have been
  cached
• Nodes act as relays when a connection to the
  internet is available
• Implements a Mail Transport Agent which
  broadcasts e-mail to peers which then forwards
  them to an SMTP server
• Doesn't take into account the possibility of
  duplicate sends

27
Peer-to-Peer in Mobile Ad-Hoc Networks

• Cross-Layer Gnutella for MANETs 24
• Uses a combination of the Gnutella protocol and
  OLSR pro-active routing protocol
• The modified Gnutella implementation communicates
  with OLSR, making it advertise the node's
  presence while receiving routing event
  information
• Abandons the idea of Super Nodes and Leaf Nodes
  as detailed in the Gnutella v.2 protocol in this
  overlay all nodes are Super Nodes
• This combination of an efficient peer-to-peer
  overlay with an established routing protocol
  results in far better performance
• Lower overheads than a non-cross-layer approach
• Better ability to cope with mobility and
  addition/removal of nodes

28
Peer-to-Peer in Mobile Ad-Hoc Networks

• Oustanding Issues
• The difficulties in resolving K.B.R with MANET
  architecture
• K.B.R works in infrastructure networks where the
  physical location of nodes is of little
  importance
• In MANETs, it is generally too costly and
  inefficient to superimpose a highly structured
  overlay onto the network which is at odds with
  the physical topology of that network
• What does this mean for DHTs in MANETs?
• Lack of agreed simulation methodology and
  validation
• ns2 and, possibly, OPNET are the de-facto
  standard tools for simulation
• Findings of the Dagstuhl Workshop on P2P MANETs
  25
• Lack of sanitized data sets for simulation
• Lack of QoS measurement tools
• Lack of agreed benchmarks for new protocols and
  simulations

29
Peer-to-Peer in Mobile Ad-Hoc Networks

• Oustanding Issues
• Resolving existing routing protocols with P2P
  unstructured overlay protocols
• Gnutella Cross-Layer implementation results in
  significantly increased network overheads
• Strict layering of unstructured overlay
  approaches on top of wireless routing protocols
  is unlikely to work
• Results in links which span multiple hops
• Increases overheads significantly for
  intermediate nodes
• Replication-based approaches work well for
  frequently-partitioned networks but requires a
  large amount of storage at intermediates
• Flooding increases network overhead and requires
  media access at each node
• Only efficient if there are a moderate number of
  nodes distributed relatively evenly

30
Peer-to-Peer in Mobile Ad-Hoc Networks

• Outstanding Issues
• Resolving existing routing protocols with
  Structured P2P Overlay protocols
• Essentially the same problem as with K.B.R and
  Unstructured protocols
• Layering an overlay on top of the MANET which is
  at odds with its physical structure results in
  increased overheads and decreased reliability
• How to combine the efficient approach of
  structured overlays with the changing topology of
  MANETs?
• O( logN ) efficiency possible in infrastructure
  networks, can a compromise protocol be devised
  which preserves this level of efficiency?
• Topology-dependant structuring represents a
  compromise which has already been implemented but
  requires object transfers on movement
• Topology-independant approaches result in
  increased lookup efficiency and don't require
  object transfers but do result in longer paths

31
Peer-to-Peer in Mobile Ad-Hoc Networks

• Questions

32
References

• Peer-to-Peer Architecture Case Study Gnutella
  Network
• Matei Ripeanu, University of Chicago, Illinois,
  U.S.A
• Characterizing Todays Gnutella Topology
• Daniel Stutzbach, Reza Rejaie, University of
  Oregon, Oregon, U.S.A
• The Bittorrent P2P File-Sharing System
  Measurements and Analysis
• J.A Pouwelse, P. Garbacki, D.H.J. Epema, H.J.
  Sips, Delft University of Technology, The
  Netherlands
• CYCLON Inexpensive Membership Management for
  Unstructured P2P Overlays
• S Voulgaris, D Gavidia, M van Steen - Journal of
  Network and Systems Management, 2005
• T-Man Gossip-based overlay topology management
• M Jelasity, O Babaoglu - Engineering
  Self-Organising Applications (ESOA05), 2005
• Squirrel A Decentralized Peer-to-Peer Web Cache
• Sitaram Iyer, Antony Rowstron and Peter Druschel,
  PODC 2002

33
References (contd.)?

• Ivy A Read/Write Peer-to-Peer File System
• Athicha Muthitacharoen, Robert Morris, Thomer M.
  Gil and Benjie Chen
• An Analysis of the Skype Peer-to-Peer Internet
  Telephony Protocol
• Salman A. Baset and Henning Schulzrinne
• Peer-to-Peer Information Retrieval Using
  Self-Organizing Semantic Overlay Networks
• Chungqiang Tang, Zhichen Xu and Sandhya Dwarkadas
• Pastry Scalable, Decentralized Object Location
  and Routing for Large-Scale Peer-to-Peer Systems
• Antony Rowstron and Peter Druschel
• Chord A Scalable Peer-to-Peer Lookup Protocol
  for Internet Applications
• Ion Stoica, Robert Morris, David Karger, M. Frans
  Kaashoek and Hari Balakrishnan
• AODV Routing Protocol Implementation Design
• Ian D. Chakeres and Elizabeth M. Belding-Royer
• A Reliability Layer for Ad-Hoc Wireless Sensor
  Network Routing
• Edoardo Biagoni and Shu Hui Chen

34
References (contd.)?

• A Review of Routing Protocols for Mobile Ad-Hoc
  Networks
• Mehran Abolhasan, Tadeusz Wysocki and Eryk
  Dutkiewicz
• GPSR Greedy Perimeter Stateless Routing for
  Wireless Networks
• Brad Karp and H. T. Kung, Harvard University
• Mobile Ad-Hoc Networking Imperatives and
  Challenges
• I. Chlamtac, M. Conti, J.J.N. Liu - Ad Hoc
  Networks, 2003
• Wireless Sensor Networks A Survey
• I.F. Akyildiz, W. Su, Y. Sankarasubramaniam and
  E. Cayirci
• Multicast Over Wireless Mobile Ad-Hoc Networks
  Present and Future Directions
• Carlos de Morais Cordeiro, Hrishikesh Gossain and
  Dharma P. Agrawal
• Adaptive Privacy-Preserving Authentication in
  Vehicular Networks
• K Sha, Y Xi, W Shi, L Schwiebert and T Zhang
• P2P MANETs New Research Issues
• M. Gerla, C. Lindemann and A. Rowstron

35
References (contd.)?

• Exploiting Epidemic Data Dissemination for
  Consistent Lookup Operations in Mobile
  Applications
• C. Lindemann and O.P. Waldhorst
• A Special Purpose Peer-to-Peer File Sharing
  System for Mobile Ad-Hoc Networks
• A. Klemm, C.Lindemann and O. Waldhorst
• 7DS - Node Cooperation and Information Exchange
  in Mostly Disconnected Networks
• S. Srinivasan, A. Moghadam, S.G. Hong and H.
  Schulzrinne
• A Cross-Layer Optimization of Gnutella for Mobile
  Ad-Hoc Networks
• M. Conti, E. Gregori and G. Turi
• P2P MANETs New Research Issues
• M. Gerla, C. Lindemann and Ant Rowstron