Federation Performance

1
Federation Performance Monitoring
Visualization Josh Bers Lynn Carlson Steve
Boswell
2
OVERVIEW

• Present tool suites for
• Collecting information on performance
  characteristics of federation machines and
  network (FedPerfMon)
• Visualizing performance data and relationships
  (FedChart)
• Along the way, discuss
• Design issues we considered
• Strengths / limitations of the approach
• Present an example of use
• Conclusions

3
MOTIVATION

• Initial goal model and predict federation
  performance
• Requires data on performance factors
• Instrument a running federation
• Organize data collection to support analysis
• Build support tools for analysis / interpretation
• Wanted generic and broadly applicable tools
• Acceptable to developers and test directors

cf. 01E-SIW-063, 01S-SIW-070, 00F-SIW-085
4
Directive Low Impact a Must

• NO code changes to use
• NO administrative work to deploy
• NO attention drain
• NO performance impacts

Developers are a tough bunch!
5
Basic Design Choices

• Sense passively whenever possible
• Use existing OS reporting services
• rperf/istatd, netstat, ntop, tcpdump
• Centralized (non-federate) collector
• generates some run-time traffic
• simplifies administrative requirements
• eg, no storage on federate machines
• Round-robin polling for machine-specific
  statistics
• Broad swath of statistics

6
BBN Federation Monitor
See 02S-SIW-038. RTIC Wed 130 pm or EMF Thurs
335pm
7
Overview of Data Collected by FedPerfMon
8
DCT / HLAresults (Logger) Post-processing

t
td
t2d
t3d
9
Use Case EnviroFed III DDM Test

• DTSim federate (HydroSim mode) publishes soil
  state
• JSAF processes soil state and calculates vehicle
  mobility
• HydroSim precalculated two conditions (dry / wet)
• HydroSim serves burst of terrain to JSAF every 3
  minutes
• Toggles between dry and wet
• Three trials No DDM, DDM-1, DDM-2 (Partitioned
  Grid)
• BBN FedPerfMon deployed
• JSAF responses and network behavior observed

10
Received Hydrogeologic Objects (w/o DDM)
Thanks to Peter M. Wickis LMIS
11
Received Hydrogeologic Objects (with DDM)
lat
Region
lon
12
Updated Terrain State (with DDM)
13
Impact of Region-based Filtering

• Without DDM, JSAF processes 238 terrain object
  updates
• With DDM, JSAF only processes 21 updates
• 91 reduction in processing load
• (CPU not network overload or swapping in this
  case)
• 90 reduction in PVD pause times
• 6 sec vs. 60 sec

14
BBN Federation Monitor
See 02S-SIW-038. RTIC Wed 130 pm or EMF Thurs
335pm
15
HydroSim Attribute Updates (10 sec Intervals)
238 HLA updates per environmental change
Update Count
16
Packets Sent by JSAF No DDM
During pauses, JSAF is not transmitting
20
18
16
14
12
packets per second
10
8
6
4
2
0
210 sec
510 sec
690 sec
870 sec
1050 sec
1002220111
1002220187
1002220254
1002220307
1002220357
1002220410
1002220460
1002220510
1002220560
1002220610
1002220660
1002220710
1002220760
1002220811
1002220860
1002220910
1002220960
1002221011
1002221060
1002221110
1002221160
1002221211
1002221260
1002221310
1002221367
Time Since Monitor Started (sec)
17
Packets Sent by JSAF DDM1
(No Space Partitioning, Single Channel)
Long pauses not observed
packets per second
18
Two DDM Configurations at IE-3

• Without space partitioning (DDM-1)
• Uses default RID file
• No multicast benefit
• Region overlap filtering performed at receiving
  LRC
• With space partitioning (DDM-2)
• 32 enumerations in subspace divided into 8
  partitions of 4 to separate HydroSim updates
• Dimensions one and two (geographic lat/long)
  divided into 5 partitions each
• 8 x 10 x 10 too hard to join
• 200 multicast groups

19
Routing Space Static Grid Partitioning
HydroUpdate
Weather
Role
or
Function
Red Air
Blue Ground
Lat

Lon
Lat

Lon
Geographical
Geographical
20
Grid ltgt Multicast Group Mapping
21
Packets Sent by JSAF DDM2
Space Partitioned, 200 Multicast Groups
Hypothesis Publication w/o Region (RTI-NG sends
to all 200 groups)
22
Network Congestion Symptom
Observed only with partitioned routing space
Ethernet collisions at HydroSim Federate
160
140
120
100
Collisions per second
80
60
40
20
0
1
6
11
16
21
26
31
36
41
46
51
56
61
66
71
76
81
86
91
96
101
106
111
116
121
10 second samples
23
FedChart Gallery of Charts
24
Multicast Traffic Plot Construction
25
Investigate Multicast Packet Explosion

• ContamSensor.Free class
• associated with breathing
• Publication w/o region. RTI-NG explodes to all
  group addresses in use.

26
Identification of HydroUpdate Multicast Group
Just before the HydroUpdates
Just after the HydroUpdates
27
Lessons and Conclusions

• FedPerfMon and FedChart as-is
• Good for troubleshooting and R/T understanding
• Quite sensitive to detail (though some questions
  are always elusive)
• Excellent planning aid
• Basis for performance quantification
• Basis for improved federate techniques
• What-ifs and preparing for new federations
• eg, JSAF turn off ContamSensor or modify
• (was needed as-is for other federates in
  EnviroFed III)
• Future Directions
• Replace round-robin polling with better snapshot
  technique
• Great opportunities for
• analytical use of collected data
• visualization aids