Transcript CRESCO SPIII.5 Project status 06-07-07

CRESCO SPIII.5
Project status 06-07-07
Dip. Di Informatica Sistemi e Produzione
Università di Roma Tor Vergata
E. Casalicchio, E.Galli, S.Tucci
Università di Roma "Tor Vergata" - CRESCO-SPIII
What's an Agent
 It's a software process able to work asynchrounsly and autonously
in an particular environment
 An agent is identifiable, a discrete individual with a set of
characteristics and rules governing its behaviors and decisionmaking capability.
 It can perceive and interact with the world (environment and
other agents)
 Agents have protocols for interaction with other agents
 Every agent is identified by an own location, capabilities and memory.
 Can adapt its behaviours according to its experience

Every agent has its own goal.
 Intelligent agent can evaluate different plans and choose the best one
2
Università di Roma "Tor Vergata" - CRESCO-SPIII
What is ABM&S
 ABMS has its direct historical roots in complex adaptive systems
(CAS)
 Systems are built from the ground-up
 Important results can emerge in systems that are completely
described by simple rules that are based on only local information
 Results that may develop can be extremely sensitive to the initial
conditions.
 Simple rules can be used to understand much of the complexity
observed in the real world
3
Università di Roma "Tor Vergata" - CRESCO-SPIII
Why ABM&S
I. Allow to simulate a complex system composed by many subsystem
organize system hierarchically
decomposition of an agent in multi-agents
simplify interactions between entities when it's to hard to
represent them with equations
II. Customize the target system as required by the case under study
III. Allow to study of unexpected pertubations on the interconnected
infrastructures
IV. Study the behaviour of an infrastructure for a certain input more
than a equations
V. Allow to give a whole picture of the system behaviour
VI. Allow to use external specific simulators, wrapping them with a
agent
4
Università di Roma "Tor Vergata" - CRESCO-SPIII
Independent or sequential simulation
IN_i
Sim i
OUT_i
Output
processing
and analysis
OUT’_i
IN_i,j
IN_j
Sim j
OUT_j
Output
processing
and analysis
OUT’_j
IN_i,k
IN_k
IN’_k
Sim k
OUT_k
Output
processing
and analysis
E.g.:
• Simulation of the effect of fault in
the power grid in [t1, t2].
• From the output i know that RomaEst city’s area will experiment a
blackout [t1’,t2’]
• I use such result to build a scenario
for the simulation of the
communication network behavior in
[t1,t2] introducing a blackout
starting at t1’ and ending at t2’
OUT’_k
5
Università di Roma "Tor Vergata" - CRESCO-SPIII
Interdependent or distributed simulation
Local input
parameters
SIM a
SIM b
RMI
RMI
o_b(t)
o_a(t)
e_i(t)
e_i(t)
GRID and/or SOA
communication
infrastructure relying
on the internet
Local Ouputs
o_d(t)
SIM d
RMI
e_i(t)
o_c(t)
e_i(t)
SIM c
RMI
o_a(t),
o_b(t)
o_c(t)
o_d(t)
Collector
RMI
Data
base
Output
processing
and
visualization
E.g.: simulation of the transportation net., the power grid and the comm. net. in Roma in [t1,t2]
• At time t’>t1 the power grid experiment a failure.
• Such failure propagates and have a direct effect at time t’+dt on a node of the comm. net.
• The failure of a comm. net. node has an impact on a not well known simulated system that directly
or indirectly use the CommNet, e.g. the transp. net.
6
Università di Roma "Tor Vergata" - CRESCO-SPIII
Advantages & Disadvantages
7
 Sequential:
 Distributed:
 -/+ macroscopic int.
 + microscopic int. analysis
analysis
 - Do not emerge unknown
interdependencies
 + Does not require
simulators re-engineering
 + could emerge unknown
Università di Roma "Tor Vergata" - CRESCO-SPIII
interdependency
 - Requires simulators reengineering or adaptation
Introduction to a custom approach with
Repast and Omnet
 We have used Repast as Toolkit for ABM&S
 We have modeled both some Infrastructure and simple users
which use them
 We have used OMNeT++ as framework DES (discrete
event simulation) to simulate Communication Network
 We have developed a distributed approach with Repast and
OMNeT to study an our defined simple scenario
8
Università di Roma "Tor Vergata" - CRESCO-SPIII
Introduction to a custom approach with
Repast and Omnet (cont.)
 In Repast has been simulated interaction between different infrastructures.
 Every Infrastructure is an agent with its behaviour, memory and so on.
 The communication network is modelled using a specialized simulator: Omnet
 The agents which communicate using the Communication networks use specific calls
of the: CommunicationNetworkAgent
 V.Cardellini, E. Casalicchio, E. Galli. Agent-based Modelling of
Interdependencies in Critical Infrastructures through UML. In Proc. of Spring
Simulation Multiconference 2007 (SpringSim07)
Wounded
IS4CEM
Wounded
R5
IS4CEM
CMN
HCC1
9
R4
Internet
R3
R1
Soccurer
Università di Roma "Tor Vergata" - CRESCOREPAST
SPIII
Soccurer
R2
OMNET
HCC1
Implementation of model:
Network
 Based on two level:
 Interactions between infrastructures are controlled by agents of Repast
 Simulation of communication network developped with OMNet++
 Developped an interface between Repast and OMNet++
 The communication network is built in OMNet++
 Setting of capacity of lynks between agents
 Simulation of workload
 The CommunicationNetwork Class of Repast control failure and
can set new capacity for a link
 Agents use CommunicationsNetwork’s method to send message
to other agents on network
10
Università di Roma "Tor Vergata" - CRESCO-SPIII
Implementation of model:
Network (cont)
 Example of sending of a message:
 Every agent that wants to send a message on network
comunicates its necessity to Communication Network Class
 Communication Network prepares messages to send and uses the
interface with OMNet++. Simulation is paused.
 OMNet++ calculates routing and time for every new message
and informs Communication Network, if any message is arrived
at destination. The Communication Network Class determine if
there is a failure in the communication (e.g. with a timeout).
 Communication Network schedules reception of messages for
every receiver-agents and the simulation restarts
11
Università di Roma "Tor Vergata" - CRESCO-SPIII
Some screenshots: Repast & Omnet
12
Università di Roma "Tor Vergata" - CRESCO-SPIII
Some screenshots: Graphs with Omnet
13
Università di Roma "Tor Vergata" - CRESCO-SPIII
Some screenshots: GUI of Omnet
14
Università di Roma "Tor Vergata" - CRESCO-SPIII
Conclusions
 Study of the common scenario
 Create an interface for every Infrastructure Simulator
 Study of interdependency between different infrastructures
 Mapping every infrastructure to a geographic location
 Use of GIS or Google's API
 Introduce a more complex fuzzy-set rules for every actors
that we want to model
15
Università di Roma "Tor Vergata" - CRESCO-SPIII