THE AGENT MECHANISM OF INTERFACE RELAXATION - PowerPoint PPT Presentation

1 / 25
About This Presentation
Title:

THE AGENT MECHANISM OF INTERFACE RELAXATION

Description:

1. THE AGENT MECHANISM OF INTERFACE RELAXATION. Giota Tsompanopoulou. Gas TurbnLab Seminar ... Use Interface Relaxation Formula to compute the new values ... – PowerPoint PPT presentation

Number of Views:17
Avg rating:3.0/5.0
Slides: 26
Provided by: giw3
Category:

less

Transcript and Presenter's Notes

Title: THE AGENT MECHANISM OF INTERFACE RELAXATION


1
THE AGENT MECHANISM OF INTERFACE RELAXATION
  • Giota Tsompanopoulou
  • Gas TurbnLab Seminar
  • March 8, 1999

2
Outline
  • PDE problem review
  • Agents
  • Communication Framework (KQML)
  • Characteristics
  • Kinds

3
PDE Review
GasTurbnLab PDE Problem
4
PDE Review (contd)
  • Solve different PDE problems
  • Combustor (KIVA-3V)
  • Compressor (ALE3D)
  • High-pressure turbine (ALE3D)
  • Low-pressure turbine (ALE3D)
  • Gas mixing (Pellpack)

5
Network of solvers
Combustor (KIVA-3V)
Gas Mixing (Pellpack)
Gas Mixing (Pellpack)
High-pressure Turbine (ALE3D)
Compressor (ALE3D)
Gas Mixing (Pellpack)
Gas Mixing (Pellpack)
Low-pressure Turbine (ALE3D)
Compressor (ALE3D)
6
Scenario
  • Define local PDE problems and their solver
    modulo.
  • Describe connectivity.
  • Set up resources.
  • Set up parameters.
  • Define action on interfaces.

7
PDEAgents
  • Characteristics
  • Kinds of Agents
  • Autonomy
  • Social ability
  • Reactivity
  • Pro-activeness
  • PDECoordinator
  • PDESolver
  • PDEMediator

8
PDECoordinator
Parse Input File
Startup the Agents
Agents
Configure Agents
Wait for messages
Global Convergence
Send/Receive messages Move to next step
Exit
9
PDECoordinator
  • Parse Input Read input filename,parse input
    file, figure out connections.
  • Startup the agents Instantiate all the other
    agents.
  • Configure the agents Send messages with
    necessary information.
  • Wait for messages
  • Exit Kills all agents and exits.

-User change of parameters, force
termination. -Agents status of
convergence. -Decide on global convergence and
initiate exit.
10
PDECoordinator
11
PDECoordinator (contd)
  • State Description
  • Parse Input File
  • Solver Configuration
  • Parse the input file
  • Get needed information ( of solvers, of
    mediators etc)

Send messages that contain
  • identification number
  • paths to the input/output data
  • of mediators to communicate with and their ids
  • kind of messages that should wait for

12
PDESolver
Agent started - Receive Data
Preprocess (.e) file - Extract boundary points
Wait on initial guess
Mediator Agents
Solve local PDE problem
Send new boundary data
Wait for instructions
No Local Convergence
Send/Receive messages Move to next step
Global Convergence
Exit
13
PDESolver
  • Agent Started (by PDECoordinator) .
  • Data Staging Receive messages to fill its model
    (from PDECoordinator). Solver id, id of
    collaborator mediators, location of input data
    and executables, etc.
  • Call Pelltool. Preprocess and compile the
    Pellpack description file.
  • Extract boundary points. Identify mesh points on
    the boundary.Send message to PDEMediators.
  • Wait for PDEMediator to compute initial data.
  • Start-supervise ExecuteTool.
  • Send new boundary data to PDEMediators.
  • Wait on convergence status messages.
  • If global convergence, plot the output.
  • If local convergence, stay idle and wait for new
    messages.


Iterate until convergence
  • Local convergence (PDEMediator).
  • Global convergence(PDECoordinator).

14
PDESolver
15
PDESolver(contd)
  • State Description
  • Extract Boundary Points
  • Legacy Program
  • Wrapper

Read the file with all mesh points (format x y
z) Pick up the points on interfaces Write them in
a file (format x y z)
Send a message to mediators that the file with
interface points is ready
16
PDEMediator
Agent started - Receive Data
Wait for boundary points
Compute initial guess - Send boundary values to
PDESolvers
Solver Agents
Wait for new boundary data
Coordinator
Apply IR method
Unsatisfied
Check convergence criteria
Send/Receive messages Move to next step
Satisfied
Exit
17
PDEMediator
  • Agent started (by PDECoordinator).
  • Data Staging Receive necessary messages (from
    PDECoordinator). Mediator id, id of the neighbor
    solvers, location of input data and executables,
    tolerance, relaxation method, initial guess, etc.
  • Wait for boundary points from PDESolvers.
  • Initialize boundary condition values.
  • Send boundary condition values to PDESolvers.
  • Wait for new boundary data.
  • Compute relative error and convergence status.
  • Send message to PDESolvers (idle, continue).
  • Send message to PDECoordinator (convergence, no
    convergence).
  • Read /combine the files with boundary points.
  • Use initial guess for boundary condition values.
  • Read output data from PDESolvers.
  • Compute the difference between two iterations.
  • Apply convergence criteria for local convergence.

18
PDEMediator
19
PDEMediator(contd)
  • State Description

Relaxation
  • Legacy Program
  • Wrapper

Read Solvers output file (format x y z u ux uy
uz) Use Interface Relaxation Formula to compute
the new values Write in its output file the new
data (format x y z u ux uy uz)
After the above computation is done, it sends a
message to left and right Solvers that the new
data are ready in the expected file
20
Send Messages. Move to next step.
Right Solver
Extract mesh/grid points on interface (from right)
Solve the local PDE problem
Read all points on interface
Read all values on interface
No
Mediator
Yes
Check Convergence Criteria (diff. of two
succ.iter., residuals, etc.)
Combine points. Give initial values
Convergence
Read the boundary values Apply IR method.
Idle/Exit
Solve the local PDE problem
No
Read all values on interface
Read all points on interface
Left Solver
Extract mesh/grid points on interface (from left)
21
KQML
  • Communication language between software agents
    (agents have a knowledge base).
  • Variety of message types.
  • Reliable message delivery.
  • Messages to single destination arrive in the
    order they were sent.
  • Non-blocking sending (sender is responsible for
    the success of sending)

22
KQML (contd)
  • Messages can not be processed in parallel.
  • Limited size of messages (actual observed limit
    2K).
  • KQML message parameters (some of them)
  • sender
  • receiver
  • content
  • reply-with
  • in-reply-to

23
KQML (contd)
  • Type of messages
  • Syntax
  • achieve (S asks R to change something in its
    environment)
  • tell (S sends to R a message from its knowledge
    base)
  • ask-one (S ask one R a question, and waits for a
    reply)
  • ask-all (S ask all Rs a question and waits for
    their replies)
  • broadcast(S wants the R to send a message to all
    possible connections)
  • reply (communicates an expected reply)

( type_of_message parameter1 value1
parameter2 value2 )
24
KQML Messages (Examples)
  • (tell sender S receiver R content
    ltexpressiongt)
  • (ask_one sender S receiver R content
    ltexpressiongt reply-with
    ltexpressiongt)
  • (reply sender S receiver R content
    ltexpressiongt in-reply-to
    ltexpressiongt)
  • (achieve sender S receiver R content
    ltexpressiongt)

25
Examples in SciAgents implementation
  • (tell receiver solver_id sender mediator_id
    content (mediator_id
    convergence case) )
  • (tell receiver solver_id sender mediator_id
    content(mediator_id bpoints mediator_id
    ) )
  • (tell receiver solver_id sender mediator_id
    content(mediator_id bvalues mediator_id
    ) )
  • (tell receiver R sender coordinator_id
    content (-1 stop) )
  • (tell receiver mediator_id sender
    coordinator_id content(-1 real_number)
    )

case 0 /1 (no convergence/convergence)
real_number new tolerance changed during run
time
Write a Comment
User Comments (0)
About PowerShow.com