Octopus (PowerPoint)

							          Octopus
A Fault-Tolerant and Efficient
  Ad-hoc Routing Protocol



                    Idit Keidar, Technion

               Joint work with Roie Melamed

 Idit Keidar, Technion    Intel Academic Seminars, February 2005   1
                         Ad-Hoc Networks
     A collection of mobile wireless nodes
     No pre-existing infrastructure
     Peer-to-peer routing: nodes relay each
      other's packets toward their ultimate
      destinations




Idit Keidar, Technion      Intel Academic Seminars, February 2005   2
 Applications of Ad-Hoc Networks
     Military: tactical communications
     Rescue missions: without adequate wireless
      coverage
     Commercial use: sales presentations
     Local Area Networks (LANs): in limited-
      coverage areas




Idit Keidar, Technion   Intel Academic Seminars, February 2005   3
    Challenges in Ad-Hoc Networks
     Lack of Infrastructure
     Limited wireless transmission range
     Rapid movement
           constantly changing topology
     Battery constrains
     Intermittent node disconnections



Idit Keidar, Technion   Intel Academic Seminars, February 2005   4
                         Multi-Hop Routing




                          A             B
                                                  C
                                                                   D

Idit Keidar, Technion        Intel Academic Seminars, February 2005   5
    Position-Based Ad-Hoc Routing
     Each node knows its location
           e.g., using GPS
     To send a packet–
           source discovers target location
           packets forwarded to this location
Knowing location can eliminate flooding,
 improve scalability



Idit Keidar, Technion   Intel Academic Seminars, February 2005   6
                         Location Severs
     Location servers for node n:
           nodes storing n’s location
           need to be updated whenever n moves
     To lookup t’s location–
           discover a location server of t
     All-for-some:
           each node has some location servers
                no flooding for update or lookup
           each node acts as location server for some nodes
           e.g., Grid Location Service (GLS) [Li et al.]

Idit Keidar, Technion       Intel Academic Seminars, February 2005   7
                         Goals and Tradeoffs
     Low location update overhead
           want to send few update packets
           do not want to send many far away (many hops)
     Fault-tolerance (overcome disconnections)
           need many location servers
           need information to be fresh (frequently updated)
     Challenge: have many fresh location servers
      without inducing high load

Idit Keidar, Technion        Intel Academic Seminars, February 2005   8
                         Observation
     In most protocols, each location update
      packet contains the location of a single node,
      and updates a single location server

     The key to a better fault-tolerance/overhead
      tradeoff is aggregation
           Challenge: locate location servers as to allow
            efficient aggregation and cheap location discovery


Idit Keidar, Technion    Intel Academic Seminars, February 2005   9
                         Octopus


Idit Keidar, Technion    Intel Academic Seminars, February 2005   10
                         Octopus in a Nutshell
     Space divided into
      horizontal and vertical
      strips
     Nodes in same strip store
      each other’s locations
     Location updates
      aggregated in each strip
     Grid can change over
      time (unlike GLS)

Idit Keidar, Technion        Intel Academic Seminars, February 2005   11
                    Octopus: Key Features
     Fault tolerant
           many fresh location servers
     Efficient
           aggregation reduces location update overhead
     Simple
     Supports dynamically changing area
     Improved forwarding


Idit Keidar, Technion    Intel Academic Seminars, February 2005   12
                         Three Sub-Protocols
     Location update
           maintains each node’s location at its designated
            location servers as well as at its radio range
            neighbors
     Location discovery
           discovers a target location (at an appropriate
            location server)
     Forwarding
           forward data packets to this location

Idit Keidar, Technion        Intel Academic Seminars, February 2005   13
Location Update I – Neighbor List
     Periodically, each node
      broadcasts HELLO
      message with its
      identity and location
           to radio-range neighbors




Idit Keidar, Technion     Intel Academic Seminars, February 2005   14
    Location Update II – End Nodes
     A north/south end
      node has no
      neighbors in
      direction north/south
      that reside in its
      vertical strip
     Same for east/west
      horizontal


Idit Keidar, Technion   Intel Academic Seminars, February 2005   15
 Location Update II – Strip Update

                         A-C      A-F A-I         A-K A-M A-P A-S        A-W




                                                                        #messages
                                                                        per node-
                                                                        constant
                                                                        # bits- sqrt
Idit Keidar, Technion         Intel Academic Seminars, February 2005                  16
               Location Discovery Take I




Idit Keidar, Technion   Intel Academic Seminars, February 2005   17
             Location Discovery Take II


        Forwarding
        Hole



          Quadratic
          reduction
          of failure rate

Idit Keidar, Technion   Intel Academic Seminars, February 2005   18
    Location Discovery Alternatives
     Two opposite directions at a time
           north and south concurrently,
           if fails, west and east concurrently
     One direction at a time
           try short direction first (use estimate of grid area)


     Tradeoff between overhead and latency


Idit Keidar, Technion    Intel Academic Seminars, February 2005    19
                 Forwarding: Geographic
     Greedy
     Forward packet
      to neighbor that
      is closest to
      target




Idit Keidar, Technion   Intel Academic Seminars, February 2005   20
             Forwarding: Local Maxima
     Geographic
      forwarding fails
     Octopus uses
      redundant
      information about
      strip nodes
     Forward to strip
      node closest to
      target
Idit Keidar, Technion   Intel Academic Seminars, February 2005   21
                                        Evaluation




Idit Keidar, Technion   Intel Academic Seminars, February 2005   22
                         NS-2 Simulations
     Scalability
           increasing the network size with fixed density
           increasing the node density
     Fault-tolerance
     Data forwarding
     Comparison with GLS




Idit Keidar, Technion      Intel Academic Seminars, February 2005   23
    Reliability: Query Success Rate




Idit Keidar, Technion   Intel Academic Seminars, February 2005   24
                         Message Complexity

                                                        Scalable!




Idit Keidar, Technion       Intel Academic Seminars, February 2005   25
                         Byte Complexity




Idit Keidar, Technion      Intel Academic Seminars, February 2005   26
              Node Density & Reliability




Idit Keidar, Technion   Intel Academic Seminars, February 2005   27
  Node Density & Message Complexity

                                                    Scalable!




Idit Keidar, Technion   Intel Academic Seminars, February 2005   28
  Node Density & Byte Complexity

                                                    Scalable!




Idit Keidar, Technion   Intel Academic Seminars, February 2005   29
                         Fault-Tolerance




Idit Keidar, Technion      Intel Academic Seminars, February 2005   31
            Data Forwarding Reliability




Idit Keidar, Technion   Intel Academic Seminars, February 2005   32
                         Comparison with GLS
     Leading solution to date
     Compare:
           Reliability
           Message and byte complexity
           Fault-tolerance
           Data forwarding reliability and overhead




Idit Keidar, Technion        Intel Academic Seminars, February 2005   33
                         Reliability




Idit Keidar, Technion   Intel Academic Seminars, February 2005   34
                         Message Complexity




Idit Keidar, Technion       Intel Academic Seminars, February 2005   35
                         Byte Complexity




Idit Keidar, Technion      Intel Academic Seminars, February 2005   36
                         Fault-Tolerance




Idit Keidar, Technion      Intel Academic Seminars, February 2005   37
                         Data Overhead




Idit Keidar, Technion     Intel Academic Seminars, February 2005   38
                     Octopus: Conclusions
     Highly fault tolerant
           reliable when all nodes intermittently disconnect
           many fresh location servers
     Efficient
           aggregates: sends much fewer messages
           saves MACs, hence sends fewer bytes
     Simple
     Supports dynamically changing area
     Forwarding uses location information

Idit Keidar, Technion    Intel Academic Seminars, February 2005   39