RealTime Performance of RTI Version 1'3

1


Real-Time Performance of RTI Version 1.3
Presented By Ernest F. Payne Raytheon Systems
2
AFRL/VACD Simulator Networking Interests

• High fidelity simulations of dynamic air vehicles
  on a network
• What does the pilot perceive across networks?
• end-to-end delays and latency
• spatial accuracy
• bandwidth
• Measurement of network performance
• understand where improvements are needed
• Improvements to efficiency of protocols
  architectures
• What are the best network architectures?

3
(No Transcript)
4
Objective of HLA Evaluation

• The purpose of the High Level Architecture (HLA)
  Evaluation is to compare its performance in
  flight simulator systems to the performance of
  the previous technology - which at the time was
  the Distributed Interactive Simulation (DIS)
  protocol

5
Approach

• Integrate RTI 1.3v2 for Silicon Graphics into the
  Network Evaluation for Training Systems (NETS)
  testing setup
• Evaluate HLA performance as a function of network
  loading
• local area and wide area end-to-end latency
• bandwidth
• accuracy

6
HLA Architecture
7
Hardware

• 2 Encore RSX machines
• 1 for each manned player
• SG Challenge machine
• 8 200 MHz R4400 processors
• E-Plex board for Ethernet connections
• IRIX 6.2
• Synchronized through external interrupts or
  shared memory flags

8
HLA Evaluation Software

• HLA Evaluation used RTI 1.3v2
• Each HLA NIU was isolated to its own CPU using
  REACT 3.0
• RPR-FOM 0.1.5
• Only data for MilitaryPlatformEntity sent
• GPS time used for dead reckoning
• Receive-order event processing
• No time regulation
• Best Effort

9
Testing Loads

• Loading was increased by increasing the number of
  digital entities and their performance
  characteristics
• 8 Trim had 4 entities on each LAN flying at trim
  condition
• Next had 1 entity on each LAN performing high-G
  maneuver
• Rest increased by 1 per side the number of
  entities performing high-G maneuver

10
Test
Thresholds Run Load Entities (deg,m,sec) Proto
col 1 8 trim 8
3,1,5 RTI 1.3 2 1v1 8
3,1,5 RTI 1.3 3 2v2 8
3,1,5 RTI 1.3 4 3v3 8 3,1,5 RTI
1.3 5 4v4 10 3,1,5 RTI
1.3 6 5v5 12 3,1,5 RTI
1.3 7 6v6 14 3,1,5 RTI
1.3 8 7v7 16 3,1,5 RTI
1.3 9 8v8 18 3,1,5 RTI
1.3 10 9v9 20 3,1,5 RTI
1.3 Table - Test
Matrix
11
Network Delay Measurements
SCRAMNet to Ethernet
Ethernet to SCRAMNet
Encore to SCRAMNet
SCRAMNet to Encore
12
Latency MeasurementsHLA vs. DIS vs. DIS-Lite
13

14
Latency MeasurementsHLA vs. DIS vs. DIS-Lite
15
Latency MeasurementsHLA vs. DIS vs. DIS-Lite
16.67
16
Bandwidth MeasurementsHLA vs. DIS vs. DIS-Lite
17
Packet MeasurementsHLA vs. DIS vs. DIS-Lite
18
Packet Efficiency
19
HLA Evaluation Results Summary

• With Respect to NETS.
• Ethernet packet inefficiencies
• Higher bandwidth consumption for equivalent
  simulations
• Higher Processing Overhead
• Increases NIU latency
• May limit number of highly-dynamic entities in a
  real-time simulation

20
HLA Lessons Learned

• Ethernet ports for data traffic can be specified
  now in 1.3v2.
• Recompilation went very smooth.
• Version Transition was easy.
• No data format of Ethernet packets on wire
  available
• SNAP would have to be intrusive to the federation
  to capture attributes from Ethernet packets
• It is desirable to have decoding functions

21
HLA Evaluation Summary

• Learned how HLA impacts network simulations
• Baselined performance of RTI 1.3v2
• Identified areas which need improvement
• Understand problems early so that research can
  have positive impact on the development of HLA
  for real time simulator applications

22
HLA Future

• 1. We would like to repeat our experiment with
  the newer versions of HLA.
• 2. We would like to measure under realistic
  simulations conditions.