Design and Implementation of a Grid Architecture over an Agent-Based Framework

1
Design and Implementation of a Grid Architecture
over an Agent-Based Framework

• A. Grosso, R. Podestà, C. Vecchiola, A.
  Boccalatte
• l.i.d.o. DIST University of Genova

speaker Alberto Grosso
2
Agenda

• Introduction
• Grids
• Agents for the Grids
• A-Grid
• Architecture (AgentService framework)
• The role of the agents
• Agents at work
• Conclusions

3
Introduction

• Grids issues
• Resource sharing
• Resource localization
• Resource storage
• User profiling
• Load balancing

4
Introduction

• Grid environment
• Complex
• Heterogeneous
• different hardware support on each node
• different nodes can host different type of
  resources
• Highly dynamic
• nodes can be added or removed
• links can be added or removed

5
Introduction

• Agents for the Grids
• Agents features
• autonomous
• reactive
• proactive
• social ability

flexible behaviour (intelligence)
6
Introduction

• Agents for the Grids
• Multi-agent systems
• social ability is one of the most important
  features of agency
• agents interact by means of
• competition
• negotiation
• cooperation
• in order to better exploit this skill agents are
  arranged in communities called multi-agent
  systems (MASs)

7
Introduction

• Agents for the Grids
• agent technology has been conceived for complex
  and highly dynamic systems
• pro-activity and reactivity become crucial in
  this kind of contexts where the single nodes need
  to adapt to
• the mutating conditions of the grid topology
• the different load during their life cycle
• social ability is important too
• cooperation is fundamental in resource discovery
• cooperation and negotiation are winning practice
  for resource acquisition

8
Introduction

• Agents for the Grids
• Some examples
• A4 methodology
• agents dynamically adapt to the conditions of the
  grid
• agents are homogeneous and organized
  hierarchically
• MyGRiD
• provides a collaborative environment for
  biologist researchers distributed in all the
  world
• uses SoFAR as implementation technology
• Bond Agent Systems
• extends the JADE framework with specific
  behaviors that abstract grid services

9
A-Grid

• Features
• Shares computational resources among nodes
• Based on the AgentService programming framework
• Uses a functional decomposition approach to apply
  agents to Grids
• Identifies all the core functionalities that
  characterize the system
• Defines a role for each set of functionalities
  that makes up a service
• Defines an agent for each role previously
  identified
• Models the entire Grid as dynamic and distributed
  multi-agent system

10
AgentService - Architecture
11
AgentService Agent Model

• Knowledge
• Shared data structure
• Concurrent access
• Deadlock avoidance
• Persistence
• Behavior
• Agent activity
• Concurrent execution
• Access to FIPA services
• Runtime environment
• Exposes platform services

12
AgentService

• Directory Service
• Platform federation
• Service agent localization
• Communication Infrastrucure
• Message exchange
• Conversations
• Inter-platfom messages
• Additional Services

13
AgentService - Communication

• Inter-platform messages
• Soap message
• Custom XML serialization
• Resource transfer
• Dedicated additional module
• FTP service

14
A-Grid

• Architecture
• Each node is defined by an instance of the
  AgentService programming platform
• The instance of the platform hosts the portion of
  the multi-agent system related to the node
• Computational node
• Access to the system

15
A-Grid

• Architecture

16
Grid Agents

• Core functionalities of the grid are provided by
  dedicated agents
• NodeManager
• Carrier
• Authenticator
• Worker

17
Grid Agents

• NodeManager
• Takes care of the list of the shared resources
• Handles the information about the topology of the
  node and neighbor nodes
• Is responsible of assigning a task to a specific
  node
• Delivers requests that it cannot handle to the
  other agents of the node

18
Grid Agents

• Carrier
• Handles the transfer of data among the nodes
• Relies on the file transfer service offered by
  the platform
• the object code containing the task
• all the related input or data files
• Interacts with the NodeManager to coordinate the
  delivery of a resource to the client

19
Grid Agents

• Authenticator
• Verifies the credential of end-user that want to
  access the grid (first level of authentication)
• Gives the user access to the specified resource
  according to its security profile (second level
  of authentication)

20
Grid Agents

• Worker
• Can be more than one for each node
• Sets up the computing environment required for
  the task
• Executes the task and eventually communicated the
  results.
• Can perform many tasks concurrently

21
Agents at work

• Computational Grid - Execute a task over the Grid
• Users that want to submit a job to the grid have
  to contact those nodes which are configured as
  access points to the grid
• These nodes are the starting point of the entire
  process

22
Agents at work

• A common computing task is submitted to the grid
• The user authenticates with by sending a message
  to the Authenticator agent of the access point
• The Authenticator agent
• checks the user permissions and validates his
  request
• forwards the request to the NodeManager
• NodeManager selects the best node and manages the
  task execution

23
Agents at work
Source Node
Target Node
Node Manager
Carrier
Carrier
Node Manager
Worker
Select target node ()
Assign (Task)
Transfer (Res,Addr)
Upload (Resources)
Execute (Task)
Ready ()
Work ()
Finished ()
Update ()
24
Agents at work

• Tasks have to implements a specific interface
• interface ITask
• bool IsReusable get
• Exception Error get
• bool Prepare(string configFile)
• void Execute()
• bool Abort()
• bool Dispose()

25
Conclusions

• Some observations
• Agent Technology can be an interesting approach
  to grid computing
• Dynamic and distributed system
• Extension and integration
• The use of agent functional decomposition leads
  to..
• .. an intuitive, modular organization of the
  system
• .. a system easy to maintain
• new approach

26
Conclusions

• Future enhancements
• Integration and interoperation with existing grid
  systems (Alchemi in particular)
• Improve the communication system and make it more
  flexible
• Extend the application contexts to other types of
  grid

27
Thanks for the attention...