SOM: Dynamic Push-Pull Channel Allocation Framework For Mobile Data Broadcasting

1
SOM Dynamic Push-Pull Channel Allocation
Framework For Mobile Data Broadcasting

• Jiun-Long Huang, Wen-Chih Peng, and Ming-Syan
  Chen
• IEEE Transactions on Mobile Computing, Vol.5,
  No.8, Aug. 2006
• Presented by Jing David Dai
• Dept. CS, VT

2
Outline

• Introduction
• State-of-art and Problem Formulation
• Analytical Models
• SOM (Solution Mapping)
• Performance Evaluation
• Conclusion

3
Introduction

• Increasing popular mobile computing environments
• Stock activities, traffic reports, weather
  forecasts,
• Wireless mobile clients
• Small batteries, limited bandwidth
• Design issue
• Conserve the energy and communication bandwidth
  of a mobile unit while allowing mobile users of
  the ability to access information from anywhere
  at anytime

4
Introduction

• Data delivery modes
• Broadcast (push)
• On-demand (pull)
• Dynamic data and channel allocation (hybrid)

5
Introduction

• Dynamic data and channel allocation
• Change data delivery configuration to achieve
  optimal performance
• If the load is heavy, more broadcast
• If the load is light, more on-demand

Lighter Load ? Heavier Load
6
Introduction

• Contributions of this paper
• Describes the analysis model of dynamic
  allocation approach
• Proposes algorithm SOM to find optimal allocation
• Devises algorithm BIS to dynamically partition
  data items and channels

7
State-of-art and Problem Formulation

• Currently not many multi-channel push and
  multi-channel pull approach
• One broadcast channel, one on-demand channel,
  fixed or dynamic data cut
• One broadcast channel, multiple on-demand
  channels
• One approach with multi-channel push and pull
  uses flat broadcast program
• Not efficient for data with different access
  probability

8
State-of-art and Problem Formulation

• System description
• n of data items (nnonb)
• no of data items in on-demand channels
• nb of data items in broadcast channels
• Ri the ith data item (0ltiltn)
• K of channels (KKoKb)
• Ko of on-demand channels
• Kb of broadcast channels

9
State-of-art and Problem Formulation

• Data dissemination

10
State-of-art and Problem Formulation

• Tasks to dynamically allocate data and channels
• Determine Ko and Kb
• Determine no and nb
• Construct hierarchical broadcast program

11
Analytical Models

• Broadcast channels
• Wb(Kb, nb) minimal average access time for data
  in broadcast channels
• C(K1, n1) the configuration that KbK1 and nbn1
• Methods to partition data to broadcast channels
• OPT can find optimal solution but time-consuming
• VFK efficiently gets the close-optimal solution

12
Analytical Models

• On-demand channels
• Wo(Ko, no) minimal average access time for data
  in on-demand channels
• Pno (no) probability that the requested data
  item is in on-demand channels as one of the no
  items
• ? request arrival rate
• ?o Pno (no) ? request arrival rate for
  on-demand channels

13
Analytical Models

• On-demand channels (Cont.)
• On-demand channels M/M/c queuing system with
  arrival rate ?o
• Service rateµ bandwidth/(data_sizerequests)
• Based on queuing theory, Wo(Ko, no)
• where

14
Analytical Models

• Overall average access time
• Probability of a requested data item in on-demand
  channel
• Average access time
• Minimal average access time

15
Analytical Models

• Trade-off of dynamic data dissemination

16
SOM (Solution Mapping)

• Problem transformation
• Find the best allocation for data items and
  channels find C(Kb, nb) with minimal W(Kb, nb),
  where 0ltKbltK and 0lt nbltn
• Search space (K1)(N1)
• SOM process
• Search space pruning phase
• Solution searching phase

17
SOM (Solution Mapping)

• Search space pruning based on following
  requirements
• nb gt Kb when 0lt Kb ltK
• nb lt n when Kb lt K
• nb 0 when Kb 0
• no 0 when Ko 0
• lt 1
• Prune effects
• using property 1-4
• Lower bound

18
SOM (Solution Mapping)

• Solution searching
• LocalOptimalCheck test a configuration is local
  optimal or not, if not, return the direction to
  local optimal
• LocalOptimalPrediction predict the position of
  local optimal based on extrapolation
• BIS process iteratively tests the unpruned
  configurations uses above two functions to find
  local optimal for all Kb finally returns the
  best solution.

19
SOM (Solution Mapping)

• BIS-Incremental
• Integrate BIS with VFK
• Since VFK is a greedy approach to find optimal
  cut point of data items, the intermediate results
  in VFK can be reused in BIS
• Each time when BIS trying to calculate Wb, it
  first check whether it has been calculated by VFK

20
SOM (Solution Mapping)

• Complexity analysis
• BIS O(K log n) time complexity of broadcast
  program
• BIS-Incremental O(K log n) 1/k Complexity of
  VFK O(log n) K (O(K log K)O(n))
• Space complexity O(Kn)

21
Performance Evaluation

• Simulation Model
• Access frequency
• ? parameter of Zipf distribution
• ? 0 uniform distribution
• Large ? skewed distribution
• Five schemes for comparison

22
Performance Evaluation

• Skewness of access frequency

23
Performance Evaluation

• Number of data items

24
Performance Evaluation

• Number of channels

25
Performance Evaluation

• Number of clients

26
Conclusion

• This paper describes the analysis model of
  multiple broadcast and on-demand channels.
• Algorithm SOM, including solution searching
  approach BIS, is proposed to find optimal
  allocations of data items and channels.
• Simulations results show the efficiency and
  scalability of SOM.
• The authors didnt address how to re-allocate the
  data items and channels.
• The quality of the optimal solution from SOM is
  not evaluated.

27
Thank you!