Large Scale Agent Based Simulation Environment

1
Large Scale Agent Based Simulation Environment

• Mahmoud Rafea, Konstantin Popov, Stelios
  Lelis, Petros Kavassalis, Fredrik Holmgren,
  Seif Haridi, Charis Lina, Jakka Sairamesh
• Swedish Institute of Computer Science, SICS,
  Box 1263, SE-16429 Kista, Sweden
• seif, kost, fredrikh, mahmoud_at_sics.se
• Institute for Computer Science, Foundation of
  Research and Technology - Hellas, ICS-FORTH,
  Science and Technology Park of Crete, P.O.Box
  1385, GR 711 10 Heraklion, Crete, Greece
• lina, slelis_at_ics.forth.gr and
  petros_at_rpcp.mit.edu
• IBM Institute for Advanced Commerce, IBM Thomas
  J. Watson Research Center, P.O. Box 704, Yorktown
  Heights, NY 10598 jramesh_at_us.ibm.com

2
Outline

• Research goals
• The simulation environment
• The model software components and their relation
  to the environment
• Specifications mechanisms to run simulation
  experiments
• An example Web Word of Mouth model
• Conclusions and future work

3
(No Transcript)
4
The Simulation Environment

• Facilitates rapid development of simulation
  models by
• Providing generic components
• Defining a methodology for implementing the model
• Allows running a series of simulation experiments
  of different model components (software) by
  providing a specification mechanism, which
  enables the configuration of the experiment from
  the implemented components.

• Experiment Environment Model components
  Specifications

• Consists of
• A control component for the sequential
  experiments
• A top-level manager for the parallel-distributed
  experiments
• A process control components
• A communication channel components for both the
  sequential and parallel experiments

5
A Metaphor

• Manager Main processor
• Process Control Cabinet processor
• Application Set of Cabinets
• Interaction/communication mechanism front panel
• Rack distribution and parallelization
  infrastructure provided by the environment

Interaction/Communication Mechanism
Manager
Rack
6
Model Components and their relation to the
environment
7
Specification Mechanisms

• Role
• Provide the simulation parameters
• Guide the control to configure the application
  (process).
• Create the desired communication channels.

• Model specifications structures for defining
• Categories one-to-many relation between category
  and components
• Components file , category
• Dependencies constraints between compoenents
• Provide one-to-many relationship between
  components and operations
• Use one-to-one relationship between channel and
  operation

• Experiment specifications structures for
  defining
• staticchannels( ltchannel gt on off) )
• dynamicchannels( ltchannel gt ch(state on
  off update integer ) )
• experiment(ltcomponent gt )
• connections(ltcomponent gtr(ltcomponent
  gtltconnectiongt )

8
Web Word of Mouth Model

• Agent behavior
• Asks friends to propose their favorite sites and
  visit them.
• Visits some sites from his portfolio.
• Surf along the links of the already visited
  sites.
• Replace a site in the portfolio by a new site if
  that new site maintained a higher utility for a
  longer period of time.

• Agent characteristics
• Have preferences for specific categories of
  content
• Participate in local social networks
• Maintain a portfolio of frequently visited sites
• Have memory about visited sites and their
  perceived utility.
• Have an evaluation method in order to evaluate
  sites they visit.

9
Conclusions and Future Work

• We have demonstrated a methodology to implement
  rapidly and efficiently an agent based simulation
  model.
• The model application is scalable through
  distribution and parallel processing.
• The work has been greatly facilitated by
  Mozart/Oz. The wonderful realm of symbolic,
  concurrent and network-transparent programming
  system.
• The behavior language development is going on in
  parallel with the improvement in the environment.
  
• Today, the parallel-distributed implementation of
  workers and agent collections components needs
  highly skillful programmers. We believe that the
  behavior language will make this task affordable
  to most of the implementers.