Ahmad Al Hanbali, Richard Boucherie, Jan-Kees van Ommeren and Roland de Haan

1
Polling systems as performance models for mobile
ad hoc networking
Ahmad Al Hanbali, Richard Boucherie, Jan-Kees van
Ommeren and Roland de Haan
2
Mobile ad hoc networks (i)

• Static and mobile hosts (e.g., laptops, cellular
  phones, PDAs)
• Wireless communications
• No infrastructure, self-organizing

3
Application (i)

• Wild-life monitoring
• Wild species equipped with radios form mobile
  ad-hoc network
• Monitoring system for interactions of individuals
  and groups

4
Application (ii)

• Vehicular networks
• Vehicles and road signs/structures form mobile
  ad-hoc network
• E.g., distribution of traffic information,
  internet access in cars

5
Application (iii)

• Disaster relief networks
• Existing communication means may fail
• Rescue workers and equipment form an ad hoc
  network to smoothen the rescue operation by
  allowing for voice/data-communication

6
Mobile ad hoc networks (ii)

• Main (functional) characteristics
• Wireless communication(dynamic channel
  conditions, interference, collisions)
• Intermittent connectivity due to
  mobility(routing is an issue)
• Distributed network control(no central entity)

7
Research questions

• Transfer delays and buffer levels in such
  networks (QoS guarantees)
• (Optimal) routing protocols
• Network design (include extra stations?)

8
Related literature

• Experimental studies(mobility process)
• Simulations
• Analytical models(infinite populations, single
  packet)

9
Our contribution

• Practice
• Analytical modelling framework for mobile ad hoc
  networks
• Theory
• Extension of the literature on polling systems

10
Research method

• Original situation
• Modelling create analytical queueing model
  containing key characteristic
• Intermittent connectivity due to mobility

11
Queueing model

• Key ingredients
• (Random) Arrival process of customers
• Service requirements
• Single server
• Simplest model M/M/1 queue
• Performance measures
• Queue length
• Sojourn time

12
Intermittent connectivity

• Assumption
• mobile stations move in local neighborhood
• Modelling
• discrete set of locations for each station
• stations remain a random time at a location
• given the specific locations of the stations,
  data transmissions may occur

13
Toy example - model

• Network 1 fixed source S, and1 mobile
  destination D

?
0
S
D
1
Depending on the location of destination D,
there exists either a link (state 0) between S
and D (i.e., S can transmit data to D) or no link
(state 1)
14
Toy example - analysis
?
S
D

• Let packets generated at S correspond to customer
  arrivals
• Let a link correspond to a server being available
  in a queueing model
• Link No
  link
• gt Specific queueing model
  Unreliable-server model Gaver, 1962

15
Towards more general networksa polling system

• Single-server multi-queue model
• Key characteristics
• Multiple queues
• Random arrival of customers
• Service requirements
• Server visits queues in some order
• Service discipline
• Performance measures
• queue-length distribution and delay measures

16
Polling system - refined

• Specific details to study MANETs
• Autonomous service discipline
• Random visit times of server ( exponential
  time-limited discipline)
• Probabilistic routing of the server
• Customer routing
• Preemptive service
• (acc. to preemptive-repeat-random policy)

17
Solution approach (i)

• Construct relations for p.g.f.s of queue-length
  distribution at specific instants, viz.
• Server arrival instants to Qi and server
  departure instants from Qi
• Server departure instants from Qi and server
  arrival instants to Qj

ai
bi
aj
bj
t
Qi
Qj
ak
bi
aj
bj
t
Qi
Qj
18
Solution approach (ii)

• System of equations (in terms of p.g.f.s)
• Numerical iterative solution method
• Joint queue-length distribution

19
Applications of polling systems (i)

• Queue-length and delay measures for, e.g.,
• Multi-hop tandem networks,
• General single-channel networks (IEEE 802.11)

?
?
S
R
D
T
?
?
?
S
U
?
?
?
V
20
Applications of polling systems (ii)

• Optimization
• Optimal delay via power control by adjusting
  visit time parameter,
• Optimal delay via channel time assignment by
  adjusting visit time parameter

?
?
S
R
D
T
?
?
?
?
S
U
?
?
?
V
21
Results

• Modelling framework for evaluation of emerging
  communication paradigms (Delay-Tolerant
  Networking, opportunistic networking)
• Exact analysis of polling systems operating under
  a novel service discipline

22
Concluding remarks

• Numerical solution approach creates need for
  (mean value) approximations
• Extend framework to
• Multi-server polling models
• More general server visit times

23
Thanks for your attention