Kimmo Raatikainen

1
Kimmo Raatikainen Chiara Braghin Stefano
Campadello Heikki Helin Oskari Koskimies Pauli
Misikangas Mikko Mäkelä Sampo Pyysalo Sasu
Tarkoma
University of Helsinki, Finland
2
Overview of Presentation

• Monads project
• Goals
• Architecture
• Predicting Quality of Service
• Other Research Issues
• Agent Communication Layers
• Wireless Java RMI
• Partitioning Applications with Agents

3
Monads Research Project

• Areas of research
• usefulness of agents in mobile computing?
• (Dis)Connection Management
• Bandwidth Optimisation
• Easy to use API
• Supporting Communication with legacy software
• agent communication in wireless environments
• adaptability to available resources
• short-term predictions of available resources

4
The Nomadic environment

• A mobile computer connected to a network with a
  wireless connection
• Characteristics of wireless networks
• low throughput, highly variable delays, sudden
  disconnections, ...
• Nomadic users need special support
• Intelligent and adaptable applications
• Optimized communication

5
Monads Adaptation Model
Adaptability
Adaptability
User Interface Agent
Service Agent
Service
Terminal Equipment
Co-operation
Data Comm. Agent
Adaptability
Communications Infrastructure
6
Monads layered Architecture
7
Monads System Services
8
Predicting QOS

• Minimal Adaptation
• Detect changes in the Quality-of-Service
• Adapt to the current QoS
• Adaptation may come too late
• Uses of Prediction
• Scheduling decisions, Data Prefetching,
  Connection Management
• Examples
• What is the expected data transfer within next
  t seconds
• What is the expected time to transfer the x KB
  of data

9
Learning to Predict QOS

• Predictions can be achieved by learning how
  certain quantities affect the QoS
• location of the terminal
• time of day
• day of week
• recent QoS values
• Division into sub-problems
• predicting terminal movement
• predicting QoS at given location and time

10
System Architecture
11
QOS Prediction - System Components

• Location Service
• Reads location information from a GPS device
• Builds a Waypoint Map
• Represents the current location as a point on a
  straight between two waypoints
• QOS Management Service
• Provides information about the current QoS
• Perception Service
• Centralized collection of observable values

12
System components (contd..)

• Route Modeler Agent
• Actively learns the regular routes of a user
• Provides predictions about user movement
• given a sequence of waypoints, which will be the
  following waypoints?
• QOS Modeler Agent
• Actively learns the characteristics of the QoS on
  the routes traveled
• Provides predictions about the QoS at given
  location and time

13
System components (contd..)

• QOS Prediction Agent
• An intelligent agent that provides QoS
  predictions
• combines predictions made by the modeler agents
• Selects between alternative ways to predict
• Informs other agents when the QoS is about to
  change significantly

14
Communication Layers

• Interaction Protocol
• FIPA-query, FIPA-request
• Message Coupling
• Communication Language
• KQML, FIPA-ACL
• Binary Encoding of Message/Content
• Message Transport
• HTTP,RMI,IIOP
• Transport Protocol
• SMS,MDCP,TCP

15
Related Research Results

• Optimized RMI
• Adaptation through Application Partitioning
• terminal characteristics
• wireless link quality