Message Ferrying Approach for Data Delivery in Sparse Mobile Ad

Document Sample
Message Ferrying Approach for Data Delivery in Sparse Mobile Ad Powered By Docstoc
					Message Ferrying Approach for
Data Delivery in Sparse Mobile
      Ad Hoc Networks
 Wenrui Zhao, Mostafa Ammar, and Ellen Zegura

                Presented by
               Justin Yackoski
The Problem – Sparse Networks
●   High chance of network not being fully
    connected at all times, but we want to be
    able to communicate to everyone all the time.
    Previous Solution – Epidemic
●   When one node encounters another node,
    give that node a copy of all stored data
    Previous Solution – Epidemic
●   Since this is a mobile network, assume
    nodes from one partition will eventually
    move close enough to another partition to
    communicate with it
    Previous Solution – Epidemic
●   The mobile node then “infects” the other
    partition with the data it knows about, so that
    the data is effectively transmitted across the
       Problems with Epidemic
●   Any ideas on problems with this?
       Problems with Epidemic
●   Will nodes ever actually move between
       Problems with Epidemic
●   How long will it take until nodes move?
       Problems with Epidemic
●   Buffering/transmitting – we must know about
    all our partition's data that needs to be
    sent to another partition, and if we want to
    send data to another partition, we have to
    tell the as many other nodes as we can so
    they can try to send it.
●   Lots of duplicate effort
    Solution – Message Ferrying
●   Some nodes (“message
    ferries”)recruited/designed specifically to
    move between partitions
    Solution – Message Ferrying
●   Introduce non-randomness into movement of
    nodes, exploit mobility of nodes to help
    deliver data.
     Why Ferry instead of fixing
●   Crisis – battlefield/disaster, existing
    connectivity destroyed, must find temporary
●   Geography – very sparse network,
    impossible to keep connected, or mountains,
    water, etc. make certain areas very sparse
●   Cost – Use existing mobile entities (buses as
    in DakNet) to reduce cost
●   Service – Provide privacy / alternate route in
    untrusted environment.
Message Ferrying Assumptions
                     (in this paper)

●   Location awareness of all nodes (GPS, etc.)

●   Single ferry

●   Ferry has unlimited buffer space to store
    messages & unlimited energy to move

●   Ferry is invincible
       Node-Initiated MF (NIMF)
●   Ferry moves through entire network on a
    well-known route.

●   Nodes pro-actively move toward ferry route
    when they want to send data to another node

●   Ferry sends out “Hello” beacons, nodes reply
    with an “echo” message, and then messages
    are exchanged. Node passes responsibility
    of delivering messages onto ferry.
Node-Initiated MF (NIMF)
        NIMF – Message Drops
●   Nodes have fixed buffers, so may drop
    messages if their buffer overflows

●   All messages have timeouts set on them,
    and will be removed from node or ferry buffer
    when they timeout
      NIMF – Trajectory Control
●   We want to minimize message drops and

●   Nodes must both move to ferry and perform
    their normal nodely duties

●   Define the Work Time Percentage (WTP) - a
    node must be free to work on its assigned
    tasks most of the time
        NIMF – Trajectory Control
●   Nodes only go to ferry if:

    –   They can fulfill their WTP requirement and go to
        the ferry

    –   The cost associated with possible message
        drops is greater than a certain threshold
        Ferry-Initiated Message
            Ferrying (FIMF)
●   NIMF uses some of the node's time and
    resources in movement to/from ferry, which
    we'd like to avoid.

●   Assuming ferry moves faster than nodes, the
    ferry can move to the nodes instead

●   Also assume nodes have a long-range radio
    to send service requests to the ferry if it is
    somewhat nearby
       Ferry-Initiated Message
           Ferrying (FIMF)
●   When a node sends a service request to the
    ferry, enters “associated” mode, and
    periodically gives ferry updated location
     FIMF – Notification Control
●   Minimize message drops while conserving use
    of energy in long-range service request
●   As in NIMF, nodes only send service requests
    when cost of not sending one (and possibly
    dropping messages) exceeds a certain
●   Only send request when ferry is less than a
    certain distance away
●   Notification Message Rate (NMR) – max
    frequency long-range messages can be sent at
      FIMF – Trajectory Control
●   Since ferry has unlimited resources, only cost
    in visiting a node is the latency caused by the
    time needed to go to and from the node.

●   Minimize total latency of visiting all nodes
    with pending service requests

●   NP-hard, so must use heuristics

●   Must estimate costs because nodes move
      FIMF – Trajectory Control
●   Nearest Neighbor (NN) – Go to closest node
    to current position with a service request,
    then to closest node from that position, etc.
    Minimize total latency of going to the nodes

●   Traffic-Aware (TA) – Consider both location
    and message drop information, try to
    minimize expected message drops (roughly
    minimizing total latency because more
    latency = more timeouts)

●   NN & TA equally good in experiments
            Experiment Setup
●   Use 802.11 DCF with 250m range for short-
    range, simplified model for long-range
●   Don't count ferry energy consumption
●   Both constant bit rate and burst traffic
Effect of Node Buffer Size on
        Delivery Rate
    Effect of Node Buffer Size on
            Delivery Rate
●   Epidemic routing needs much larger buffer to
    have similar message delivery rate because
    of buffer contention in storing other nodes'
Effect of Node Buffer Size on
       Message Delay
    Effect of Node Buffer Size on
           Message Delay
●   Delay in epidemic routing is lower because
    MF explicitly delays going to ferry if there is
    plenty of room in the buffer

●   Also need to consider that MF delivers 2X as
    many messages as epidemic, so some
    messages do get through faster, others don't
    make it at all
    Effect of Node Buffer Size on
          Energy Efficiency
●   =
    Effect of Node Buffer Size on
          Energy Efficiency
●   MF only needs 2 hops (source to ferry, ferry
    to receiver) to deliver message

●   MF avoids flooding

●   With larger buffer, FIMF can broadcast long-
    range service request less frequently

●   NIMF not affected, however cost of physically
    moving to ferry is not considered
Effect of Node Mobility on
    Energy Efficiency
      Effect of Node Mobility on
          Energy Efficiency
●   Epidemic routing works better with increased
    mobility, because more nodes are infected
    with a message faster, increasing its delivery
    rate & reducing delay

●   FIMF is not affected because the ferry pro
    actively moves to nodes

●   NIMF is only affected by node speed, since
    nodes are the ones moving to the ferry
Effect of WTP on NIMF
Effect of WTP on NIMF
Effect of WTP on NIMF
        Effect of WTP on NIMF
●   Increase in WTP causes decrease in delivery
    rate because nodes are allowed to visit ferry
    less often

●   Increase in WTP does not affect message
    delay because messages are already
    queued as long as buffer allows

●   Increase in WTP lowers energy efficiency
    because energy is wasted on messages that
    end up timing out in the ferry
Effect of NMR on FIMF
Effect of NMR on FIMF
Effect of NMR on FIMF
         Effect of NMR on FIMF
●   Increase in NMR only affects delivery rate if
    timeout is small

●   Increase in NMR does not change message
    delay because message transmissions
    already delayed as long as possible

●   Increase in NMR increases energy efficiency
    because more messages are batched up
    before a service request is sent out
Effect of Transmission Range
    on FIMF Performance
Effect of Transmission Range
    on FIMF Performance
Effect of Transmission Range
    on FIMF Performance
    Effect of Transmission Range
        on FIMF Performance
●   Increase in range increases delivery rate
    because ferry route is shorter, so less
    timeouts on ferry

●   Increase in range does not affect message
    delay due to batching done by nodes

●   Increase in range causes some increase in
    energy efficiency to the extent it reduces
    drops. For large range, node energy is
    wasted because ferry is not moving as much
●   MF is better than epidemic routing at
    providing a high delivery rate and a high
    energy efficiency. Message delay is higher,
    but could be lowered as needed if energy
    efficiency relaxed.
                Future Work
●   Multiple ferries – ferries need to cooperate
    and determine who will service which
    requests, areas, etc. Ferries could also
    exchange messages with eachother

●   Node coordination – nodes could forward
    other nodes' messages a short number of
    hops, a “gateway” node could inform
    neighbors of ferry's presence and allow
    relaying of messages to ferry
                Future Work
●   Long-Range Communication – There are
    situations where using long-rage
    communication to directly deliver message
    between nodes is better (i.e., faster) than
    using the ferry depending on the application.
              The Questions
●   Give a situation/constraint where either NIMF
    or FIMF is superior over the other & briefly
    explain why

●   How could the ferry and destination contact
    each other to finish delivery of messages in
    NIMF or FIMF? Paper is not explicit on this
●   Zhao W., Ammar M., and Zegura E. A
    Message Ferrying Approach for Data
    Delivery in Sparse Mobile Ad Hoc Networks.
    MobiHoc '04. May 24-26, 2004, Roppongi,

Shared By: