01ESIW039 Paper

1
RTI Performance in the Distributed Real-time
Vehicle Model Simulation in a 3-D Graphical
Environment
Paper Number 01E-SIW-039 25-27 June 2001
Michihiko OGATA TOSHIBA CorporationTokyo,
JAPAN81-(0)44-548-5832miki.ogata_at_toshiba.co.jp
2
Why are we doing this?

• Situation
• We built a prototype of the real-time detailed
  model simulation in a 3-D graphical environment
• However,
• The performance of our simulation depends on the
  performance of RTI
• Need to examine and evaluate the performance of
  RTI

3
Overview

• About our simulation
• What is our vehicle model simulation?
• Architecture
• Performance of RTI using our models
• Objective of performance
• Test and evaluation
• Conclusion and future plans

4
About our simulation
5
General characteristics

• Detailed real-time vehicle models
• Situational awareness
• Command and control
• Decision making
• Route planning
• Distributed environment
• HLA interface
• HCI (Human computer interface)
• 2D/3D situation display

6
Architecture
7
Federation architecture
8
Situation displays
9
Performance of RTI using our models
10
Objective of performance

• Keep real-time processing
• with real-time 3-D situation desplay
• Data refresh rate
• 30Hz and above
• Examine and evaluate with DIS and RTI-NGs
• Data refresh rates
• RTI time delay
• CPU loads
• Joining time
• Stability

11
Situations and solutions

• Situation
• Data refresh rate was 30Hz and above with DIS
• It fell to 1Hz using RTI time management service
  with DMSO RTI-NGv1
• Solution
• Built our original time management system with
  wall clock instead of using RTI time management
  service
• Used periodical time synchronization between
  federates by NTP (Network Time Protocol) and
  asynchronous operation with each federate

12
Case 1 DIS - RTI-NGv1

• Experience
• Data refresh rate rose to around 30Hz
• Problems
• Cyclic delay for data refresh occurred
• Federates were hard to join to RTI with heavy CPU
  loads
• RTI was unstable

13
Case 2 RTI-NGv1 - RTI-NGv2

• Experience
• Cyclic time delay still occurred
• RTI was still unstable with federate joined

14
Case 3 RTI-NGv2 - RTI-NGv3

• Experience
• Data refresh rate became stable
• RTI was also stable
• Federates were easy to join to RTI
• We cleared our objective with performance

15
Case 4 RTI-NGv3 - RTI-NGv3.1

• Experience
• Data refresh rate was stable
• RTI was also stable
• Federates were easy to join to RTI
• We cleared our objective with performance

16
Case 5 RTI-NGv3.1 - RTI-NGv3.2

• Experience
• Data refresh rate was stable
• RTI was also stable
• Federates were easy to join to RTI
• We cleared our objective with performance

17
RTI Time Delay
() Test Condition Using DMSO Benchmark test
condition Number of federates 2 federate
Number of objects 1 object/federate Number
of attributes 1 attribute/object Size of
attributes 128 bytes/attribute Number of
updates 1000 times
18
CPU loads
19
Joining time (1/2)

• Case 1 DMSO RTI-NGv1
• It took 3 to 5 minutes
• The more federates that tried to join to RTI, the
  more joining time was needed
• Case 2 DMSO RTI-NGv2
• It took about 2 minutes
• The more federates that tried to join to RTI, the
  more joining time was needed
• Case 3 DMSO RTI-NGv3
• It took about 1 minute
• Joining time was stable

20
Joining time (2/2)

• Case 4 DMSO RTI-NGv3.1
• It took about 1 minute
• Joining time was stable
• Case 5 DMSO RTI-NGv3.2
• It took about 1 minute
• Joining time was stable

21
Stability (1/2)

• Case 1 DMSO RTI-NGv1
• Some Federates failed to join
• One federate could not recognize other federates
  most of the time
• Case 2 DMSO RTI-NGv2
• Some Federates failed to join
• One federate could not recognize other federates
  sometimes
• Case 3 DMSO RTI-NGv3
• All federates could join easily
• No problems were found

22
Stability (2/2)

• Case 4 DMSO RTI-NGv3.1
• All federates could join easily
• No problems were found
• Case 5 DMSO RTI-NGv3.2
• All federates could join easily
• No problems were found

23
Conclusion and future plans
24
Conclusion

• Experience
• Tried kinds of versions of RTI-NG
• Not found any great difference among NGv3,
  NGv3.1, and NGv3.2
• Problems about test condition
• Minimum time of measure for benchmark test was 5
  ms
• Windows clock mechanism proved to be slightly
  inaccurate
• Lessons learned
• valuable lessons regarding our real-time
  simulation

25
Future plans

• Continue to evaluate the most up-to-date version
  of RTI using our models
• Continue to evaluate our simulation with the most
  up-to-date version of RTI
• Several issues remain unsolved
• Work-in-progress

26
We hope that

• Our study find a better environment for real-time
  detailed model simulation

27
Any Questions?
28
See you again on the next SIW!Thanks a lot.
29
Contacts
Michihiko OGATA TOSHIBA Corporation, Komukai
OperationsNetwork Systems Engineering Dept.1
Komukai-Toshiba-cho, Saiwai-kuKawasaki, Kanagawa
212-8581JAPANPhone 81-(0)44-548-5832Facsimile
81-(0)44-548-5947E-mail miki.ogata_at_toshiba.co
.jp
Akira HIGASHIDEakira.higashide_at_toshiba.co.jp
Toshinao TAKAGItakagi_at_yk.toshiba-tesco.co.jp
Kenji IMAIimai_at_yk.toshiba-tesco.co.jp