Knowledgedriven Interactions with Services Across Ad Hoc Networks

1
Knowledge-driven Interactions withServices
Across Ad Hoc Networks

• Rohan Sen, Radu Handorean, Gruia-Catalin Roman,
  and Gregory Hackmann
• Mobile Computing Laboratory
• Department of Computer Science and Engineering

2
Outline

• Background
• Motivating example
• Problem description
• Accommodating knowledge in SOC
• Software architecture
• Implementation details
• Conclusions
• Future directions

3
Background Ad Hoc Networks

• Description
• No fixed network infrastructure
• Network supported by member hosts
• Resource poor, heterogeneous devices
• Characteristics
• Dynamic network topology
• Frequent disconnections leading to transient
  connectivity
• Peer-to-peer interactions

4
Background A SOC Architecture

• Adaptation of the Jini model
• Proxy-based
• Service selection using Java interfaces
  polymorphism
• Key differences
• No leasing mechanism
• Distributed service directories
• Proxies able to process certain requests locally
• Tuple space communication
• Advantages of the approach
• Abstraction of client-server protocol
• Lightweight client application

5
Motivating Example
Requirement
Services
Time
? Requirement Interval ?
6
Problem Description

• Intelligent decisions are required when choosing
  services
• Choose the best candidate from a set of services
• The best candidate is described as the service
  which
• Has the capabilities and attributes desired
• Conforms to policies
• Is available when initially required
• Is available for the entire duration of the
  requirement

7
Solution Use Knowledge

• What is knowledge?
• Meta-data that describes non-functional behavior
  of a service
• Knowledge-driven service view
• Service Advertisement Capabilities Attributes
  Service Knowledge
• Key idea Make already available implicit
  knowledge explicit

Capabilities
Attributes
Knowledge
Inherited
Service
Service Advertisement
8
Accommodating Knowledge in SOC

• Services advertise themselves in terms of
• Their capabilities and attributes
• Associated and inherited knowledge
• Focus on spatiotemporal knowledge
• Clients specify a service requirement profile
• List of services and required capabilities
• Time interval when the services are likely to be
  used
• Challenge Calculate a satisfying set of services
• That fulfills the service requirement profile
• Obeys all spatiotemporal constraints

9
Calculating the Satisfying Set

• Consider all known services
• Eliminate services that do not meet capability
  requirements
• For remaining services check that
• They are following its advertised profile
• They are going to be available and within range
• Choose the service that best conforms to the
  policy
• Repeat for every service in the requirement
  profile

10
Software Architecture
Traditional SOC Interface
Knowledge Managed Service Directory
Knowledge Manager
Aggregator
Disseminator
Planner
Manager
Manager
Storage
Storage
Communication Handler
Communication Handler
Communication Knowledge
Communication Middleware
11
Exchanging Knowledge Among Hosts

• The Knowledge Disseminator
• Makes knowledge pertaining to the local host
  available to others
• Parameter level granularity
• Single or repeated disseminations
• The Knowledge Aggregator
• Accumulates knowledge about other hosts by
  gossiping
• On-demand on a host by host basis
• Parameter level granularity
• The Planner
• Implementors of desired policies

12
Impact of Knowledge
Client
Knowledge Manager
Discover
Service
Service Directory
Updates
Updates
Communication Middleware
Knowledge Updates
13
Implementation

• Uses LIME as communication middleware for ad hoc
  networks
• Service directories implemented as LIME tuple
  spaces
• LIMEs polymorphic matching responsible for
  service selection
• Services returned as executable proxies

14
Implementation

• Knowledge made available in the form of tuples

• Knowledge obtained by registering reactions

Knowledge
Knowledge
Location
Location
Any Type
23.42N, 65.23W
Any Location
Neptune
Neptune
Any Host
15
Conclusions

• Knowledge about how hosts behave in an ad hoc
  network can help increase the predictability of
  interactions
• Make already existing implicit knowledge explicit
• We developed
• A structured specification of the problem
• An algorithm that represents one possible
  solution
• A software architecture to handle knowledge
  exchange
• An implementation targeted towards ad hoc
  networks
• However much more needs to be done

16
Future Work

• Technical challenges
• Planning when only partial knowledge is available
• Handling incorrect knowledge
• Knowledge propagation over disconnected routes
• Analysis
• Patterns of knowledge propagation
• Probability of success under different sets of
  assumptions
• Dynamic lengths of interaction between client and
  service

17
Thank You!

• Rohan Sen
• rohan.sen_at_wustl.edu
• www.cse.wustl.edu/ras6
• Mobile Computing Laboratory
• mobilab.cse.wustl.edu

18
ImplementationTransiently Shared Tuple Spaces
Tuple Spaces Merging
Tuple Spaces Merged
Tuple Spaces Disjoint