Performance Evaluation of Time-based and Hop-based TTL Schemes in Partially Connected Ad Hoc Neworks

1
Performance Evaluation of Time-based and
Hop-based TTL Schemes in Partially Connected Ad
Hoc Neworks

• Wing Ho Yuen
• and
• Henning Schulzrinne
• Department of Computer Science
• Columbia University

2
Application Driven Networking Research

• Networking research community has tradition of
  focusing on topics that find sparse applications
• E.g. QoS, multicast, active networks
• Most recent advances in networking driven by
  applications
• Email in 80s, Web in 90s, p2p and voip in 00s
• The playfield
• Cellular data is ubiquitous, but fairly expensive
• Wi-Fi available only in limited locations
• What niche should ad hoc networks fits in?

3
7DS Application

• Enable communications beyond Wi-Fi locations
  using ad hoc networks
• Node density insufficient to maintain connected
  networks in many scenarios
• Instead of dropping packets, revert to message
  based communication
• To emulate Internet services when users are
  disconnected
• Email delivery, web access and novel applications
  such as localization

4
3 Categories of 7DS Application
Subway map download
P2p music exchanges
Email delivery
5
Email Delivery Application

• Message replication speeds up email delivery
• Message purge mechanism removes replicas
• Time-based Scheme (TB)
• Each message has a TTL field
• Replicas purged at all nodes when TTL expires
• Hop-based Scheme (HB)
• Each message carries of TTL tuple (b,h)
• b decrements every time a node transmits
• When b0, message replica is purged
• When h0, a node does not Tx in a node encounter

6
Hop-based Scheme HB(b3,h2)
3,2
2,2
1,1?0,0
1,2
2,1
0,2
3,1
3,1
3,0
3,1
3,0
3,0
3,0
3,0
3,0
3,0
7
Availability of Feedback Channel

• Notify all nodes upon delivery
• Exploit paging channel of cellular network
• Replicas at all nodes immediately purged when a
  replica reaches an AP
• Four schemes under examination
• TB
• HB
• TB/FB
• HB/FB

8
HB/FB (b3,h2)
3,2
2,2
1,1?0,0
1,2
2,1
0,2
3,1
3,1
3,0
3,1
9
System Model and Performance Metrics

• Delivery of one email message
• Xn message replicas at time n
• ? node encounter rate is Poisson
• ?n ?(1-Xn/n) is effective node encounter rate
• ? AP encounter rate is Poisson
• n number of 7DS users
• ?n?? holds when Xnn
• Storage/Energy cost C
• Storage-time cost S

10
Contributions and Sim Setup

• Derive ECTB ,VarCTB, ECHB and VarSHB for
  non-feedback schemes
• Simulation results only for feedback schemes
• Each scenario corresponds to a given node
  encounter rate ? and AP encounter rate ?
• 3 scenarios node encounter is constant, ?1/20
• vary ?, with ?1/1000, ?1/2000, ?1/5000
• Consider schemes with prob. Delivery Pd ? 1
• No physical node mobility
• Node encounters generated non-homogeneous
  Poisson process ?n

11
Comparison of TB and HB
TB inferior to all HB schemes for both
storageand storage-time cost CHB decreases with
bmax SHB increases with bmax
12
Comparison of TB/FB and HB/FB
CTB dramatically reduces with FB CHB decreases
with bmax, good to be aggressive
Similar storage-time cost STB and SHB
13
Expected Stopping time ETs

• Defined as time when all replicas are purged

WITH no FB,
TB trades off small ETs for more storage Not as
meaningful as storage-time cost
WITH FB,

• stopping time message delivery time
• Message delivery in 100sec for all schemes
• ETs of HB similar to TB, at much smaller cost

14
Conclusion

• HB(bmax,2) superior, exhibit smaller expected
  cost and small variance (storage cost and
  storage-time cost)
• bmax should be selected such that Pd ? 1
• Expedited (100sec) and reliable message delivery
  (Pd 1) possible at moderate resource overhead
  (50 replicas at ?1/1000, i.e. 17 minutes to AP)
• Feedback reduces cost further by two or three
  times

15
Questions?