A Partition Detection System for Mobile Adhoc Networks

1
A Partition Detection System for Mobile Ad-hoc
Networks

• Hartmut Ritter, Rolf Winter, Jochen Schiller
• 2004 SECON
• Daekyeong Moon

2
Motivation

• Challenge
• In mobile ad-hoc networks, partitioning is
  frequent due to the mobility
• To restart an entire network due to (temporal)
  partitioning is too expensive
• Observations
• Not all partitioning is harmful
• Applications can benefit from the partition
  detection
• e.g.
• Loading entire network is sometimes too
  expensive.
• Released address space can be reclaimed.
• More precise topology control is possible.
• Problem of previous researches
• Cant distinguish node failures from
  partitioning.

3
Algorithm

• Border node (Active Node)
• A node with relatively small of neighbors.
• Border nodes are likely to be at the border of
  networks and to experience partitioning.
• Key idea
• Border nodes exchange beacon messages among each
  other.
• Further btw nodes, larger portion of network can
  be covered.
• Border node election
• Each node periodically compares its neighbor
  count with threshold and determine whether to be
  a border node or not to be.
• Two adjacent nodes cant be border nodes at the
  same time
• Threshold computation
• Static method Predetermined one. Not precise
• Dynamic method
• A node piggybacks its neighbor count onto
  application data packet.
• Each node can learn the average of the entire
  network by eavesdropping

4
Algorithm (contd)

• Buddy node
• A one-hop neighbor of a border node
• Shares the information about the peer border
  nodes with a corresponding border node
• Key idea
• Buddy nodes check border nodes up/down to
  prevent false partitioning alarm
• Ping the border node
• If the border node goes down, the buddies notify
  the peer borders
• Selecting buddy nodes.
• By listening message exchanging of the border
  node
• Buddy nodes can be changed due to the mobility
• Partition detection
• If a border cannot hear its peer, partitioning
  can be suspected.
• If no notification from buddy of suspected border
  node, regard as a partitioning.
• Otherwise, not partition but the peer borders
  failure

5
Key Parameters

• Threshold to be a border node
• Large value increases the of border nodes.
  Thus, coverage is improved. But increases
  complexities (i.e. memory , communication, )
• Frequency of a borders beacon message
• Small value improves detection latency, but
  increases communication overhead
• Frequency of buddies ping message
• The authors insist its impact is negligible
  because of one-hop message. But Really?
• Congestion Radio contention

6
Experiment (Using NS2)

• Size 50 nodes. Each node moves randomly
• Buddys ping interval 5 sec
• Nodes border election interval 15 sec
• Running time 500 sec

7
Experiment (contd)

• Totally, 3121 pkts during 500 sec
• 0.125 pkts / sec / node

8
Conclusion

• Contributions
• Presents a light-weight partition detecting
  system
• Distinguishes node failures from partitioning by
  buddy nodes. (i.e. more precise detection)
• Drawbacks
• Long distance is preferred for large coverage,
  but more hops results in more packet drops and
  finally false alarm.
• Didnt present detection latency result (only
  traffic result)

9
(No Transcript)