Minimum Disruption Service Composition and Recovery over Mobile Ad Hoc Networks

1
Minimum Disruption Service Composition and
Recovery over Mobile Ad Hoc Networks

• Shanshan Jiang, Yuan Xue, and
• Douglas C. Schmidt

Institute for Software Integrated
Systems Department of EECS Vanderbilt
University MOBIQUITOUS 2007, PhiladelphiaAugust
8th, 2007
2
OUTLINE

• Background Introduction and Motivation
• Minimum Disruption Service Composition and
  Recovery Problem Formulation
• Optimal and Heuristic Solutions
• Simulation Study

3
BACKGROUND

• Dynamic Nature of Mobile Ad Hoc Network
• It is a self-organized network of mobile nodes
  connected by wireless links.
• It can be deployed rapidly without the support of
  any fixed networking infrastructure.

• Component-based Software System
• Component-based software development has been
  used to ensure the flexibility and
  maintainability of software systems.
• Service Composition integrates loosely coupled
  distributed service components into a composite
  service to provide comprehensive functions for
  end users.
• Service Composition technique in mobile ad hoc
  network could satisfy diverse application needs
  in ubiquitous computing environments.

Our work studies the Service Composition issue in
mobile ad hoc networks.
4
NETWORK AND SERVICE MODEL
Example of Service Composition in a Mobile Ad Hoc
Network
Service Link an overlay link that may consists
of several wireless links
Service Path a linked sequence of service
components
Service Composition the process of finding a
service path in the network from many of their
replicas
5
SERVICE COMPOSITION OVER MOBILE AD HOC NETWORKS

• Existing research
• Extensive research has been done on service
  composition techniques over wire-line networks to
  satisfy QoS requirements and provide highly
  available services.
• However, they cannot be extended directly to
  service composition in mobile ad hoc networks due
  to dynamic link connectivity and network
  topology.
• There also has been extensive research on
  achieving reliable data delivery or reliable
  routing in mobile ad hoc networks.
• However, little existing work has considered
  service deliveries spanning multiple components.

Service Composition over Mobile Ad Hoc Networks
is still an open issue.
6
SERVICE COMPOSITION AND RECOVERY FRAMEWORK FOR
MANET
Service routing, which selects the service
components out of many replicas for the service
path.
Network routing, which finds the network path
that connects the selected service components.
Understand the relation between service routing
and network routing.
7
SERVICE DISRUPTION

• Service Disruption
• Due to the dynamic nature of Mobile Ad Hoc
  Networks, the service is unavailable to end users
  during the service failure and recovery
  processes, thereby causing service disruptions.
• Service availability
• Service availability is a commonly used metric
  that quantify the service delivery ability in a
  system.
• However, it is insufficient to evaluate the
  effect of user-perceived disruptions since it can
  not characterize the impact of disruption
  frequency or duration.

Quantitatively characterize the impact of
user-perceived service disruptions.
8
SERVICE DISRUPTION MODEL
9
PROBLEM FORMULATION AND SOLUTIONS

• Minimum Disruption Service Composition and
  Recovery (MDSCR) problem

• Solutions to the MDSCR problem
• Optimal solution based on dynamic programming
• Heuristic solution based on one-step look-ahead
  approximation and service link lifetime
  prediction

10
MDSCR HEURISTIC SOLUTION
We present a one-step look-ahead approximation
method where future disruption index is estimated
in the time period until its first service
failure. Let be the estimated
service level minimum disruption index from time
instance tw.
sustainability of the new service path
recovery duration from the failed service path to
the new service path
11
SIMULATION STUDY (1/6)

• MDSCR algorithm
• SPSCR (shortest path service composition and
  recovery) algorithm
• Extension of SP routing algorithm, where the
  service path with the shortest service link will
  be selected

Default simulation parameters
12
SIMULATION STUDY (2/6)
Throughput comparison with SPSCR and MDSCR
13
SIMULATION STUDY (3/6)
Disruption improvement ratio under default
simulation parameters (service path length of 4)
14
SIMULATION STUDY (4/6)
Improvement ratio comparison with concave,
linear, and convex penalty function F
15
SIMULATION STUDY (5/6)
Impact of node pause time on improvement ratio
Impact of node speed on improvement ratio
16
SIMULATION STUDY (6/6)
Impact of service link length requirement H on
improvement ratio
Impact of number of component replicas on
improvement ratio
17
CONCLUSION

• A theoretical framework for service composition
  and recovery strategies for MANETs that
  characterize the effect of service disruption

• An optimal solution to MDSCR problem based on
  dynamic programming techniques and provides
  important analytical insights for MDSCR heuristic
  algorithm design

• A simple and effective statistical model based on
  linear regression that predicts the lifetime of a
  service link in the presence of highly correlated
  wireless link failures and the network path
  repairs

18
THANKS!
19
MDSCR OPTIMAL SOLUTION
MDSCR is essentially a dynamic programming
problem. Let be the minimum
disruption index from time instance tw when
composition scheme is used.
Based on the definition of service disruption
index, we have
the minimal summary of all the following
disruption durations
20
MDSCR OPTIMAL SOLUTION
MDSCR is essentially a dynamic programming
problem. Let be the minimum
disruption index from time instance tw when
composition scheme is used.
Based on the definition of service disruption
index, we have
Based on dynamic programming, we have
Service-level MDSCR problem
21
MDSCR HEURISTIC SOLUTION
tw
tw1
tw
tw1
t
We present a one-step look-ahead approximation
method where future disruption index is estimated
in the time period until its first service
failure. Let be the estimated
service level minimum disruption index from time
instance tw.
approximated number of component substitutions
estimated failure rate
expected lifetime for service link
22
LIFETIME PREDICTION BASED ON LINEAL REGRESSION
Calculate the predicted future distance between
two components, and use linear regression to
estimate the service link lifetime.
In the simulation study, we pick the prediction
time ?t with the smallest standard error and
derive the corresponding parameters for linear
regression.