HLABased Distributed Simulation of Queueing Network Models

1
HLA-Based Distributed Simulation of Queueing
Network Models

• Johannes Lüthi
• Gerald Anders
• Institut für Technische Informatik
• Universität der Bundeswehr München

2
Background and Motivation

• Who ?
• Why ?

3
Background

• Student-project (with 6 CS students)
• FT distributed simulation with the HLA
• Goals
• HLA-based testbed of a fault tolerant
  distributed simulator, focusing on
  reconfiguration
• Performance and dependability models of
  FT distributed simulation
• Two main subprojects
• Resource Sharing System (Lüthi/Großmann, PADS
  2001)
• Distributed Simulation of Queueing Network Models

4
Resource Sharing System (RSS)

• Simulation with the HLA
• Motivation
• Architecture

5
MotivationDistributed Simulation with the HLA
Node 3
HLA Run Time Infrastructure (RTI)
Node 1
Node 2
Federation
New Node 4 ?
6
Motivation for Resource Sharing System (RSS)

• Dynamic exploitation of computing resources for
  distributed event simulation
• Users of Workstations should be able to control
  availability of their computers for the
  distributed simulation run
• Possibility to use the Resource Sharing System
  (RSS) for reconfiguration management in fault
  tolerant distributed HLA simulation
• Existing approaches (e.g. Condor) not feasable
• ? RSS Lüthi/Großmann, PADS 2001

7
HLA / RSS Architecture
8
Motivation for HLA-Based DDES of QNMs

• Needed test case for RSS FT Simulation
• Wanted HLA simulation model with
• scalability w.r.t. number of federates
• scalability w.r.t. size of federates
• control on communication granularity
• DDES of Queueing Network Models (QNMs)
• QNMs can become very large -gt DDES!
• Desired properties OK
• Partitioning of existing model into submodels
  instead of combining existing submodels

9
HLA-Based Distributed Simulation of QNMs

• HLA/RSS QNM Simulation Architecture
• Federation Editor
• QNM Federate

10
HLA/RSSQNM Simulation Architecture
Node 1
Federation Editor (using QNM specification
syntax)
Fed. Start
QNMFed 2
11
Federation Editor (FedEdit)QNM Description Syntax

• KEYWORDS
• Federate
• Separates submodels (federates)
• Node
• Queue
• Capacity, strategy, content, min time in queue
• Server
• Service time, jobs in, jobs out, successors
• Currently defined timing variables
• Fixed, exponential, uniform

12
(No Transcript)
13
Federation Editor (FedEdit)User Interface

• ASCII files
• Syntax highlighting
• Allows comments
• Syntax check
• Automatically generate data files for all
  QNMFeds(C data structures)
• Automatically start replicated QNMFeds

14
QNM Federate - Data Structures
15
QNM Federate - Synchronization

• Conservative Synchronization Chandy/Misra 79
• Supplied via RTI
• After simulation of events with timestamp t
• Request timeAdvanceGrant to timestamp t of
  next event
• Lookahead
• Crucial factor in conservative simulation
• Determines amount of parallelism
• Only nodes producing external events considered
• Improved lookahead computation subject to future
  work

16
Lessons Learned andOngoing work

• Federation execution management
• Fault Tolerance

17
Need for Federation Execution Management

• For timeAdvanceGrant decisions the RTI can only
  take into acount federates that are registered at
  the RTI!
• At simulation start not all QNMFeds can register
  simultaneously!
• Actual simulation start has to be delayed until
  all QNMFeds are registered
• QNMFeds know total number of federates
• During RSS migration a migrating federate is
  temporaly not registered in the federation!
• RSS blocks simulation appropriately

18
Ongoing and Future Work

• Federation Execution Management
• Synchronized startup and shutdown
• Support for repeatability of experiments
• Statistical support
• Configurable Fault Tolerance
• Classification of failure/fault types
• Classification of FT menchanisms
• Reconfiguration management (based on RSS)
• Fault Tolerance Based on Replication
• Berchtold/Hezel ESM 2001

19
Thanks for your Open Attention ! Questions ?