Phrenology Forum

1
2000 Fall SimulationInteroperability
WorkshopOutbrief Session
2
Federation Performance Virtual Forum

• Duncan Miller, Chair

Total attendance 120 Papers and
presentations 5 Panel discussion on fed_perf
consortium
3
Goals for this Workshop

• Discuss metrics for characterizing
• Federation performance
• Federate performance
• Run-Time Infrastructure performance
• Discuss scalability of performance with
  federation size and complexity
• Consider lessons learned from the
  implementation, instrumentation, and analysis of
  the performance of diverse federations

4
Presentations

• 00F-SIW-116 Designing FOMs Robert Murrayfor
  Performance
• 00F-SIW-085 Towards Predictive Stephen
  KolekModels of Federation Performance
• 00F-SIW-009 Scalability RTI-NG LaTreva Pounds
• 00F-SIW-031 RTI-NG Performance Wayne Civinskasin
  Large-Scale Platform-LevelFederations
• 00F-SIW-041 Optimal Usage of Thomas
  UslanderCORBA for Soft Real-time Simulation
  Applications

5
Fed_Perf Consortium

• Members agree to share data and models with the
  Consortium and report the results of their
  federation performance modeling efforts at SIWs
  and other appropriate forums
• Members may port Consortium-developed tools to
  any computing platform/OS combination
• Any such ports shall be placed in a Consortium
  repository (with documentation) for use by other
  members on an as-is basis
• Members may extend Consortium-developed tools to
  support specific applications
• Developers are encouraged to document extensions
  and place them in the repository for use on an
  as-is basis
• The Consortium cannot promise that any such
  extensions will be supported in future releases
  of Consortium tools

http//dspmo.ll.mit.edu email
info_at_dspmo.ll.mit.edu
6
Key Points

• Different federations may have very different
  performance goals (minimum latency,
  maximumthroughput, maximum runs per hour, )
• Predictive models of federation performance
  require
• Modeling the communications infrastructure
• Modeling the RTI being used
• Modeling federate service call statistics
• Performance modeling is an iterative process
• Develop tools and useful metrics collect data
• Develop quantitative performance models
• Compare predicted and measured performance
• Improve tools and models repeat as necessary

7
Recommendations

• Continue Fed_Perf as a Virtual Forum
• Develop common performance metrics for use in
  different categories of applications
• Encourage sharing of quantitative performance
  models drawn from different MS applications
• Encourage tool development and sharing on an
  open-source basis