SDSR

1
SDSR Superior DSR

• Jay Chen
• Siddharth Gidwani
• Christopher Yap

2
Mobile Ad-Hoc Networks - MANETS

• Mobile Ad-Hoc Networks
• A collection of wireless mobile nodes forming a
  communication network without centralized
  administration or existing network infrastructure
• Nodes perform both host and routing duties in
  order to transmit packets across the network
• Applications
• Wireless computing
• Sensor networks
• Security infrastructure
• Military
• Search and rescue operations

3
MANET Routing Protocol Overview

• Destination Sequenced Distance Vector (DSDV)
• Nodes maintain next-hop information
• Periodic broadcast routing updates
• Temporally Ordered Routing Algorithm (TORA)
• Broadcast-determined routes errors backflow
  packets
• Periodically transmitted heartbeats to maintain
  neighbor list
• Ad-Hoc On-Demand Distance Vector (AODV)
• Uses on-demand DSR route discovery/maintenance
• Hop by hop routing and periodic beacons like DSDV
• Dynamic Source Routing (DSR)
• Packets carry the entire hop by hop source route
  to the destination
• No setup overhead, everything done on-demand (no
  periodic broadcasts for neighbor connectivity)

4
MANET Routing Protocols - Continued

• MANET protocols present a spectrum of choices
  varying from on-demand routing to shortest-path
  routing (periodic updates)
• Turns out DSR is the most popular of the lot, as
  it offers superior performance under common
  applications and various deployment scenarios
• A major reason DSR is preferred in MANETS is due
  to the elimination of the overhead from
  periodically updating state

5
DSR Routing Mechanisms

• Route discovery
• DSR nodes perform a flooding route request, which
  is propagated through the network
• Eventually it will be propagated to a node that
  knows a path to the destination (ultimately the
  destination node)
• This node returns a route reply containing the
  source route to the requested destination
• Route maintenance
• Should a sent packet be unable to progress due to
  link failure, a route error is generated and
  propagated back to the sender
• The sender will then resort to either other
  cached routes, or perform a route discovery as a
  last resort

6
DSR - Weakness

• Even though route requests are on demand, each
  request is a propagated broadcast
• Results in unnecessary congestion both from the
  route requests and the associated route replies

7
The Approach

• Considerations
• Trend of mobile computing is that of more
  powerful nodes
• This is the not the case in sensor networks
• Idea
• Add some hierarchy
• Reduce route flooding by limiting route request
  functionality to a subset of nodes
• Interaction with the proxy should be unicast

8
Related Work - Spine Routing

• Spine Routing
• Adds a level of hierarchy to the routing
• Select a set of nodes to perform the bulk of the
  routing operations
• Non-spine nodes communicate through spine nodes
  in order to interact with the network
• Exactly what we wanted, but
• Optimal spine networks require global knowledge
  of the topology very difficult in MANETs where
  nodes are mobile

9
The Goal

• Maintain the demand-based attractiveness of DSR
  and leverage the advantages of Spine Routing
  without the overhead of maintaining a highly
  consistent spine
• Maintaining some extra state at each node to lift
  some network overhead caused by flooding route
  requests

10
SDSR Protocol Description

• Selective Dynamic Source Routing (SDSR)
• Maintain the on-demand advantages of DSR, well
  maintain a weakly consistent spine that is also
  demand determined
• Elect a few nodes in the network to serve as
  proxies
• These proxies perform route request on behalf of
  their clients
• Re-election of proxies as needed
• All other operations fall back on standard DSR
• Our lower bound for routing overhead if all nodes
  act as proxies is the same as DSR

11
Implementing Testing the Protocol

• NS-2 network simulator
• Modified the DSR implementation available in ns-2
  to add the concept of proxies
• Traffic model
• We generate simple all pairs random traffic
  scenarios
• Mobility model
• Research area in itself
• Random waypoint
• Random direction
• Terrain mappings
• Structured group mobility
• Flocking/swarming groups
• We spend some effort implementing interesting
  and more realistic mobility models

12
Mobility Models

• Random waypoint
• Most previous work done on this topic involved
  the random waypoint model and it is a point of
  basis for validating results
• Structured group mobility
• Probably of use to military/disaster recovery
  efforts
• May be a good fit with SDSR versus DSR since each
  group operates in close proximity with each other
  and we can reduce the out of group traffic to a
  single node ideally

13
Simulation Parameters

• Simulation Area
• 500m x 500m
• Number of Nodes
• 10, 25, 50 nodes
• Motion uniform/non-uniform speed
• 1-20, 20-25, 1-50, 20-50 m/s
• Pause selection constant/uniform
• 1, 10 s
• Number of connections
• 1, 10, 100 connections/s
• TCP connections instead of CBR (constant bit
  rate) since CBR seems to be geared more towards
  sensor networks

14
Structured Group Mobility Parameters

• Group Mobility
• Number of groups
• 1-4
• 10 nodes per group
• Kept every parameter constant and varied one

15
iNSpect - Visualization tool
16
Performance Metrics

• Routing overhead
• Route request packets sent/forwarded
• Route replies sent/forwarded
• Route availability
• Performance ratio defined as the number of resend
  attempts on data packets vs the number actually
  sent
• Note that a single packet may be attempted to be
  resent many times or zero times before it is
  actually sent out
• This is essentially a measure of the standard
  concept of availability negatively weighted by
  the duration of each un-available route

17
Performance Metrics - Continued

• Route lengths
• We expect to be worse, since for non-proxy nodes
  the routes will contain an extra hop
  corresponding to the proxies

18
Performance Results
19
(No Transcript)
20
(No Transcript)
21
Preliminary Results

• Cases where we do worse
• Cases where we do better
• Group mobility model
• Parameter tweaking

22
SDSR Flexibility Parameter tweaking

• SDSR performance is tied to five very important
  parameters that are set
• Proxy request timeout (client)
• Maximum time to wait for a get proxy response
  before declaring itself as the proxy
• Get request timeout (client)
• Time to wait for a proxy route response before
  attempting to find a new proxy
• Client list timeout (proxy)
• Time to wait before purging a non-communicating
  client
• Client threshold (proxy)
• Number of clients proxy must maintain to continue
  being a proxy
• Initial proxy timeout (proxy)
• Grace period for new proxies before being
  subjected to the client threshold

23
SDSR Flexibility Parameter tweaking

• By tweaking these parameters we can greatly
  influence the performance of our protocol
• Preliminary experimentation shows that we can
  improve the performance of certain test cases
  over DSR
• A point of future work might be to investigate
  determining these parameters dynamically, in
  order to optimally fit the situation at hand

24
Conclusion

• Random waypoint model
• 50 fewer routing related messages in some cases
• Structured group mobility model
• Differences in performance are minimal
• Parameter tweaking
• Control knob for tradeoff between individual node
  load and network congestion

25
Future work

• Microbenchmarks of actual CPU load and memory
  usage
• Not possible in ns-2
• More runs for statistical significance of our
  test simulations
• Investigation of other topologies and mobility
  models
• Actual implementation and testing

26
Questions?

• Questions?