Localized Operations for Distributed minimum energy multicast algorithm in mobile ad hoc networks

1
Localized Operations for Distributed minimum
energy multicast algorithm in mobile ad hoc
networks

• Paper by Song Guo and Oliver Yang supporting
  images and definitions from Wikipedia
• Presentation prepared by Al Funk, VT CS 6204,
  10/30/07

2
Table of Contents

• Background and related work
• Models system, network, mobility
• DMEM algorithm
• Operations
• Performance
• Conclusions

3
Background and related work

• Multicast communication technique which enables
  a source to send a single packet to reach
  multiple receivers.
• Objective Create a distributed algorithm to
  solve the Minimum Energy Multicast (MEM) problem
• Definition of MEM Find a route for multicast
  transmission with the minimum total energy
  consumption for a given communication session.
• Challenges MANET changing network topology, lack
  of central authority problem is NP-hard

4
Background and related work

• Prior research focused on
• Creating centralized, not distributed, algorithms
• Efficient heuristic algorithm design
• Weaknesses of prior research
• Examination of static, not dynamic, network
  topologies
• Little examination of performance impact of node
  mobility

5
Models System Model

• Discrete Power Level Management Model
• Transmission range based on power level, but
  power level increases at an exponential rate as
  distance increases
• Identify discrete power levels appropriate to
  reach nodes at various distances from the
  transmitter
• Vary transmitter power in granular increments to
  balance power use with the bandwidth usage
  necessary to constantly adjust transmission
  strength

6
Models System Model

• Pvu Power level required to transmit from node
  v to node u
• lvu Layer (concentric ring from prior slide) of
  u relative to v
• K Number of discrete power levels of the
  transmitter (and therefore number of layers)
• rK Distance of ring K
• a Parameter (2 to 4) representing rate of
  signal attenuation

7
Models Network Model

• Represent network as a directed graph, G(N,A,p)
• N set of nodes, A set of arcs, p function
  representing power required for each arc
• Rooted tree directed acyclic graph with a source
  node that has no incoming arcs and where other
  nodes have a single incoming arc
• Leaf vs. internal/relay nodes

8
Models Network Model

• For any node v in the rooted tree,there exists a
  single acyclic source route pv
• Our goal is to set lv, the transmission layer of
  node v, to the minimum necessary for v to reach
  all of its child nodes
• Once this is known, we can calculate pv, the
  necessary power level for the node

9
Models Mobility

• Mobility is a differentiator for the
  contribution, as alternative models require the
  significant overhead associated with central
  coordination.
• Authors use Random Waypoint Model
• Calculate random speeds bounded by Vmin and Vmax
  assume random start and end points introduce
  pause between journeys.
• Objective calculate the steady-state average
  speed

10
Algorithm Data Structure

• We need to store the forwarding state at each
  tree node v.
• Membership status sender, receiver, forwarder
  (can be receiver and forwarder)
• Source route p directed path from the source to
  node v (used to avoid loops)
• Tree neighborhood table TNv stores neighbors,
  along with whether is a father, child or other,
  along with layer lvu

11
Algorithm Tree Construction

• Minimum Spanning Tree Given a connected,
  undirected graph with weighted edges, an MST is a
  subgraph which connects all vertices together
  resulting in the minimum total weight.

12
Algorithm Tree Construction

• MULTICAST-JOIN-REQUEST (MJREQ) Broadcast
  message initiated by the source used when no
  route information is known
• MULTICAST-JOIN-REPLY (MJREP)Response message
  sent to previous hop node
• MJREQ Transmitted at maximum transmission power
• MJREP Returned at necessary power
• Necessary power determined by strength of the
  original MJREQ message

13
Algorithm Tree Flood

• MULTICAST-ALIVE (MA) Message sent periodically
  during session to refresh the tree (otherwise
  tree routes are cleared)
• Message sent at maximum power
• Used to adjust power dynamically
• Only sent if received from father (but then
  always sent)
• Supports tree repair and energy saving operations
• Nodes update neighborhood information to identify
  nearby nodes

14
Localized Operations

• Normal Energy Saving (NES) Upon receipt of MA
  from children, node adjusts its transmission
  power to the minimum necessary.
• Reactive approach which could lower total power
  utilization
• Keeps the tree connected but not with maximum
  efficiency

15
Localized Operations SHO

• Soft Hand-Off (SHO) Initiated by a node that
  detects it is leaving its fathers transmission
  range (K).
• Goal is to identify a new father s.t.and power
  utilization is minimized
• Node severs link with previous father (via
  MULTICAST-LEAVE (ML) message), selects the new
  father
• Tree is maintained.

16
Localized Operations MTR

• Multicast Tree Repair (MTR) In the case where
  loss of a node results in a tree partition, we
  need a way to repair the multicast tree.
• Occurs when a forwarder or receiver fails to
  receive successive MAs from its father
• Nodes furthest from the source attempt to
  reconnect first
• MULTICAST-JOIN-SOLICITATION (MJS) Hop-limited
  message

17
Localized Operations MTR

• Disconnected node closest to source notifies the
  subtree that it is initiating repair procedures
  using an MA message
• The closest node to the source initiates an MJREP
  message and attempts to reconnect the subtree
  back to the multicast tree
• If an appropriate node responds, the tree is
  reconnected if not, other nodes in the subtree
  attempt to reconnect, and the node(s) that failed
  must rejoin through a network flood.

18
Localized Operations AES

• Advanced Energy Savings (AES) A proactive method
  of reallocating child nodes s.t. overall power
  utilization of the system is reduced.
• The major contribution of the paper
• We must be able to retain the MST structure for
  multicast
• Operation performed as part of MA
• Approach Each child node attempts to extend its
  transmission range to become the parent of a
  current child of its father but only if such a
  change reduces the total power utilization of the
  system
• More sophisticated than NES
• They are not mutually exclusive

19
Localized Operations AES

• Using the MA message header means that no
  separate message is necessary for the operation
• Use of MA messages fits the algorithm -- father
  to child propagation enables communication of
  power levels and supports child decision-making.
• At each transmission from its father, a node
  modifies header with its own information and
  propagates to its neighbors
• Because MA messages are at full power, neighbors
  of multicast tree nodes will receive.
• As a result, non-multicast tree nodes can join,
  but must consider potential added cost of the
  link from a father node

20
Localized Operations AES

• AES-REQUEST When a node identifies a power
  savings, it sends an AES-REQUEST to the source
• Source reviews AES-REQUEST messages and sends
  AES-REPLY to the node with the greatest power
  savings

21
Localized Operations AES

• Finalizing the update
• Selected node sends local broadcast TREE-UPDATE
  and assigns itself as father to the node to move
• Moving node leaves father, sending
  MULTICAST-LEAVE.
• If selected node is a non-tree node, it must find
  a father
• It will be a forwarding node only, otherwise it
  would have been part of the original tree
• Multiple nodes may become children of the
  selected node if power savings justify

22
Localized Operations AES

• Examples of AES tree revision

23
Performance Evaluation

• Simulations
• Ad hoc network with size 1,000 meters sq.
• Each node can transmit 250 meters
• K10
• a 2
• Modeled max node movement speeds of1, 5, 10,
  15, 20 and 25 m/s
• Multicast groups 5, 25, 50, 75, 100
• Static networks considered
• 50 scenarios for each multicast group

24
Performance Evaluation

• Measures
• Relative tree power Ratio of actual total tree
  power for heuristic algorithm vs. ideal of MST
  algorithm
• Average tree power Power used over time for the
  tree
• Communication overheads Overhead for AES, SHO
  and MTR as a total number of these operations
  over each simulation

25
Performance Evaluation

• Static network evaluation
• Compared DMEM against prior work
• Not key to the paper, but demonstrates that DMEM
  is a useful heuristic compared with prior
  research

26
Performance Evaluation

• Mobile network evaluation Consider with and
  without optional protocol components

27
Performance Evaluation

• Examine AES performance considering node speed
  and multicast group size.

28
Performance Evaluation

• Examine SHO operations given node speed and
  multicast group size.

29
Performance Evaluation

• MTR operations considering node speed and
  multicast group size.

30
Conclusions

• In a static network, DMEM is superior to
  alternative algorithms for medium and large
  multicast groups.
• Measures heuristics, but major contribution is on
  dynamic network
• DMEM is efficient in reducing energy utilization
• AES provides significant value relative to base
  case
• SHO is mostly redundant when using AES
• DMEM proven correct for maintaining tree
  structure using localized operations

31
Critique

• Graphs are not presented in such a way to
  visually support the analysis
• e.g., authors require visual comparison of
  separate charts to compare AES and SHO, rather
  than presenting a single chart
• Is it scalable? Authors indicate that AES
  becomes saturated this seems to occur rapidly in
  large networks even at slow speed.
• Authors indicate that it is scalable with regard
  to mobility but AES saturation seems to put
  this in question, as do some of their comments
  right before the conclusion
• If scalability is an issue, possible approaches
  to address it would have been welcome
• Do the arbitration performed by the source node
  along with the broadcast approach amount to
  centralization that reduces scalability and
  creates a bottleneck?