On Demand Multicast Routing Protocol

1
On Demand Multicast Routing Protocol

• ECE 802/609
• Sameer Arora
• ODMRP in Multihop Wireless Mobile Networks, Lee,
  Su, Gerla, Mobile Networks and Applications, 2002

2
Salient Features

• Mesh-based
• On Demand route discovery maintenance
• Soft state group membership maintenance
• Does not rely on any underlying unicast protocol

3
Mesh Creation

• Source initiates by broadcasting JOIN QUERY
• JOIN QUERY (JQ) has Src ID and Seq
• Intermediate nodes
• Cache Src ID Seq from JQ in routing table -
  to detect duplicates
• Re-broadcast JQ
• Receiver nodes create broadcast JOIN REPLY (JR)
• JR has an entry of source and next hop node ID to
  Source (i.e. node forwarding the JQ)
• Single JR packet may have multiple entries
  corresp. to JQs from multiple intermediate nodes

4
Mesh Creation contd.

• On receiving JR, an intermediate node
• Checks if any next hope node ID matches its own
• If so, concludes its on the path to Source, sets
  flag FG_FLAG
• This process constructs and updates routes from
  Sources to Receivers and creates a mesh
  Forwarding Group (FG)
• FG has route redundancy, which avoids
  reconfigurations due to
• Node displacements
• Channel fading

5
JOIN REPLY Reliability

• Reliable JR transmission to source critical for
  establishment and refresh of multicast routes
• However, standard MAC protocols support reliable
  transmission of unicast (through ACK), not
  broadcast
• Two approaches
• Unicast to use MACs reliable transmission
• Broadcast with passive acknowledgements

6
JR Reliability - Unicast

• Subdivide JR
• One sub-table for each distinct next-hop node
• Unicast JR to each next-hop node
• Not scalable, but works well for optimum number
  of neighbors (6)
• Can be used as backup to Broadcast JR

7
JR Reliability - Broadcast

• When B forwards As packet to C, A can listen
• A may still miss Passive ACKs
• if out of Bs propagation range
• suffers Hidden Terminal
• If A timeouts on Passive ACKs, re-transmit
• If re-transmissions remain un-ACKed
• invalidate route
• Broadcast looking for next-hop to Src
• Neighbors with route to Src, unicast JR to
  next-hop, sets FG_FLAG
• Neighbor without route, broadcasts its ignorance,
  sets FG_FLAG
• Causes excessive redundancy, but expires on next
  JR propagation

8
ODMRP Working

• Once forwarding groups and routes are
  established
• Source can multicast via selected routes and
  forwarding groups
• A node forwards multicast packets if not
  duplicate and FG_FLAG has not expired
• Soft state timers
• Route-refresh timeout interval determines
  frequency of JQ
• Small interval ? More recent route membership
  info ? congestion
• Big interval ? Stale route membership info ?
  Unsuitable for highly mobile networks
• Forwarding Group timeout interval determines
  FG_FLAG expiration
• High traffic Small interval ? Quickly demote
  unnecessary nodes ? Reduce excessive redundancy
• High mobility Big interval ? More alternative
  routes available
• Soft State
• To leave, Source stops sending JQs
• To leave, Receiver stops sending JRs ? FG_FLAG
  expires for forwarding nodes lying on receivers
  path to source ? route expires

9
Data structures

• Route table
• Created on-demand, maintained by every node
• Forwarding group table
• Maintained by forwarding group node
• Message Cache
• Maintained by every node to detect duplicates
• For JQ or data packets
• Cache empties FIFO/LRU may be employed

10
Mobility Prediction

• Uses GPS, NTP
• Based on motion parameters location, speed,
  direction
• JQ carry motion parameters, MIN_LET (Link
  expiration time).
• Source sets MIN_LET to MAX_LET_VALUE in initial
  JQ
• A forwarding node predicts LET between itself
  sender node ? sets MIN_LET to min predicted
  received LET? substitutes its own motion info?
  forwards JQ
• Receiver predicts LET between itself forwarding
  node ? sets RET (Route Expiration Time) in JR ?
  broadcast JR
• On receiving multiple JRs, forwarding node
  selects min RET ? broadcasts JR
• Source sets refresh interval to min RET of all
  JRs ? flood new JQ before min RET expires ?
  Routes preserved in mobility

11
Mobility Prediction

• Such predicted RET may be
• Too small due to high mobility ? leads to
  excessive flooding
• Too large due to low mobility ? long refresh
  interval ? leads to long intervals between JQs ?
  refresh problems
• Distant non-members have to wait long to receive
  JQ
• If a node suddenly changes direction/speed, route
  breaks before RET ends ? route would not be
  constructed in time.
• So, RET is selected between MIN_REFRESH_INTERVAL
  and MAX_REFRESH_INTERVAL
• Values of these intervals should be adaptive to
  network environments
• Alternative mobility prediction
• When GPS may not work properly ( indoor, fading)
• Based on detected signal strengths of received
  packets

12
Simulation setup

• GloMoSim
• 50 mobile nodes, 1000m x 1000m
• Channel capacity 2 Mbps
• Average no of neighbors/node 6.82
• Free space propagation model, no n/w partitions
• 802.11 MAC
• DCF mode for broadcast (e.g. for JQs)
• RTS/CTS for unicast (e.g. for JRs)
• Evaluated AMRIS, CAMP, AMRoute, flooding in same
  setup
• No mobility prediction, 512 bytes data payload
• Multicast members stay from beginning to end

13
Metrics

• Packet delivery ratio packets delivered
• packets supposed to be delivered
• Measures protocol effectiveness
• Data packets transmitted
• data packet delivered
• Data packets transmitted every packet
  transmitted by every node over entire network
  (including drops retransmissions)
• Unlike in unicast, this can be smaller than 1
• No. of controlheader bytes transmitted
• delivered data byte
• Measures utilization of control packets in
  delivering data
• No. of control and data packets transmitted
• delivered data packet
• Measures efficiency in terms of channel access

14
Packet delivery ration vs. Mobility Speed

• Pre-defined speeds 0 72 km/h, 20 members, 5
  sources, 2pkts/sec
• ODMRP performs good even in high mobility - due
  to Mesh topology
• CAMP suffers - packets to distant routers not
  delivered due to fewer redundant paths away from
  core
• AMRIS suffers due lack to redundant paths b/w
  any member pairs
• AMRoute worst in high mobility due to loops
  and sub-optimal trees
• Relies on underlying unicast protocol to set
  bi-directional links

15
Data Packets TXed /delivered data packet

• Mesh-based protocols transmit more data packets
  than AMRIS, which tree-based
• ODMRP transmits nearly as much as flooding

16
No. of Control Bytes TXed/Data byte

• Flooding has no control packets. Pkt-header
  overhead doesnt increase with mobility
• AMRIS low control overhead because transmits
  less data in the first place
• CAMPs control overhead overtakes ODMRPs with
  high mobility because underlying unicast protocol
  causes triggered updates
• AMRoute high control overhead due to loops and
  inefficient trees

17
No. of Packets TXed/data packet delivered

• AMRIS smallest ratio because tree-based
• AMRoute highest ratio because of loops
• CAMP has smaller number of transmission than
  ODMRP
• ODMR transmits more data on redundant paths than
  CAMP
• Even though CAM has more control bytes,, control
  packets for updates are piggybacked onto updates
  of WRP (underlying DV based unicast protocol)

18
Packet Deliver ratio vs. No of Senders (Group
size constant)

• Performance of flooding slightly degrades with
  increase in senders due to collision, congestion
• For ODMRP CAMP performance remains robust,
• in fact improves because better path redundancy
  increasing number of senders
• For AMRIS, AMRoute performance unaffected
  because they use shared tree

19
Packet Delivery Ratio vs.. Group Size

• ODMRP performance unaffected by group size
• CAMP performance improves with group size
  bigger group ? more redundant paths for nodes
  away from core
• AMRIS improves but not much due to lack of
  redundant paths
• AMRoute ratio drops since one-way critical
  links increase as tree grows

20
Packet Delivery Ratio vs.. Traffic

• ODMRP better than flooding because no of data pkt
  transmissions is less than flooding
• ODMRP better than CAMP due to less control
  overhead and suffers less buffer overflow than
  CAMP
• AMRoute suffers buffer overflow at tree-members
  and mesh nodes connecting trees
• AMRIS suffers high collisions because of size and
  transmission of beacon packets

21
Summary of ODMRP

• Mesh-based protocol - robust to high mobility
• Low channel and storage overhead due to lack of
  control packets on link breaks
• even in highly mobile networks
• On-demand soft state approach to membership
  maintenance
• Usage of up-to-date shortest routes
• Reliable construction of routes and forwarding
  groups
• JQ Flooding may contribute to congestion
  contention, collision

22
Summary of Protocols