Self-Organizing Hierarchical Routing for Scalable Ad Hoc Networking

1
Self-Organizing Hierarchical Routing for Scalable
Ad Hoc Networking

• David B. Johnson
• Department of Computer ScienceRice University
• dbj_at_cs.rice.edu

Monarch Project
2
Introduction

• Safari project goals
• Self-organizing, adaptive network hierarchy
• Scalable ad hoc network routing (10s of thousands
  of nodes)
• Self-organizing higher layer network services and
  applications
• Integrated with Internet infrastructure where it
  exists
• Safari leverages and tightly integrates two areas
  of research
• Ad hoc networking
• Peer-to-peer networking
• Builds an adaptive, proximity-based hierarchy of
  cells and
• leverages this for scalable routing and higher
  layer services
• Funded by NSF Special Projects in Networking
  Research
• (January 2004)

3
Safari Hierarchy Self-Organization

• All nodes are equivalent no specialized nodes
  assumed

4
Safari Hierarchy Self-Organization

• Nodes self-elect to become a buoy

5
Safari Hierarchy Self-Organization

• Buoy nodes send limited propagation beacon floods

6
Safari Hierarchy Self-Organization

• Other nodes associate with a buoy to form cells

7
Safari Hierarchy Self-Organization

• Buoys at one level self-elect to become buoy at
  next higher level

8
Safari Hierarchy Self-Organization

• Forming cells at each higher level too

9
Simulation Example
10
Safari Coordinates

• A nodes coordinates associated cell id at each
  hierarchy level

11
Safari Routing Overview

• Destination node coordinates
• Stored and looked up in Distributed Hash Table
  (DHT) using embedded peer-to-peer system
• Hybrid routing protocol components
• Route to destination cell following beacons
  (proactive routing)
• Incremental local repair in this path (reactive
  routing)
• Route to destination node within final cell
  (reactive routing)
• Routing table at a node
• Remembers information from beacons received
• Coordinates of buoy sending beacon
• Previous hop node from which beacon received
• Hop count back to the buoy
• Sequence number of most recent beacon from that
  buoy

12
Proactive Inter-cell Routing

• Range of beacons from a buoy node
• Nodes in the cell associated with that buoy
• Nodes a few hops away, giving them a chance to
  join that cell
• Nodes in the containing cell one level up in the
  hierarchy
• Routing table lookup algorithm
• Nodes outside the cell hear the beacons
• Reasons described above
• Wireless propagation allows nearby nodes to hear
  too
• Longest common prefix matching (similar to
  Internet !)
• Compare your own coordinates to each entry in
  routing table
• As soon as packet comes to node with more
  detailed table entry, packet starts following
  lower in routing hierarchy
• Packets are routed toward buoys, not through
  buoys!

13
Routing Example

• Source node S is sending a packet to destination
  node D

S
D
14
Routing Example

• Follow beacon path toward level 3 cell in which D
  is located

S
D
15
Routing Example

• Follow beacon path toward level 2 cell in which D
  is located

S
D
16
Routing Example

• Follow beacon path toward level 1 cell in which D
  is located

S
D
17
Reactive Intra-cell Routing

• Dynamic Source Routing protocol (DSR)
• Discovers routes only as needed, on demand (Route
  Discovery)
• Detects when links being used for routing are
  broken, on demand only as they are used (Route
  Maintenance)
• Very low overhead, scalable to mobility and
  traffic needs
• Zero overhead until new route is needed
• Using DSR in Safari routing
• DSR originally designed for small or medium sized
  networks
• Safari intended to scale to much larger sizes
• Safari uses DSR only within destination
  fundamental cell
• Size of fundamental cells created by Safari
  balance two things
• Small enough for very easy efficient reactive
  routing
• Large enough to minimize when nodes move to new
  cells

18
Routing Example

• On-demand DSR routing to destination node D

S
D
19
Reactive Inter-cell Route Repair

• Beacons are sent only periodically
• Long interval between beacons important for low
  overhead
• The higher the level in hierarchy, the less
  frequent the beacon
• Following beacon reverse path may fail if nodes
  have moved
• Safari local route repair in the
• beacon paths
• Limited-hop on-demandRoute Discovery
• Flood flows downhill withlimited uphill
  allowed
• Altitude is ? prefix lengthmatched, sequence
  , hop count ?
• Result reestablishes new pathas if original
  beacon path

Increasing altitudewith hops awayfrom buoy
A node hasmoved awayfrom buoy
Buoy node
20
Simulation Evaluation

• Simulated using ns-2, includes detailed physical
  model
• IEEE 802.11 at 2 Mbps, nominal range 250 m
• Studied scale from 50 to 1000 nodes
• Randomly distributed in space
• Density maintained equivalent to 50 nodes in
  1000?1000 m
• Studied percentage of nodes being mobile from 0
  to 100
• Moving with Random Waypoint model, average 5 m/s
• Data traffic is Constant Bit Rate (CBR)
• Flows with randomly chosen source and destination
• 4 packets/second, 64 bytes/packet
• Metrics shown
• Packet Delivery Ratio percentage of packets
  delivered
• Overhead individual transmissions of routing
  packets

21
PDR vs. Number of Nodes
22
PDR vs. Percentage of Mobile Nodes
(1000 nodes total)
23
Overhead vs. Number of Nodes
24
Overhead vs. Percentage of Mobile Nodes
(1000 nodes total)
25
Conclusion

• Safari is highly scalable and provides a basis
  for services
• Forms an adaptive, proximity-based hierarchy of
  cells
• PDR and routing overhead change little with scale
  or mobility
• Performance studied through both simulation and
  analysis
• Ongoing and future work
• Further optimization and evaluation of beaconing,
  cell membership, routing, local repair
• Interconnection to traditional Internet
  infrastructure
• Higher layer services exploiting the hierarchy
  and P2P
• Testbed and experimentation