Parallel Simulation of LargeScale Heterogeneous Communication Systems

1
Parallel Simulation of Large-Scale Heterogeneous
Communication Systems
Rajive Bagrodia Julian HsuScalable Simulation
Solutions Mineo Takai Computer Science
Department, UCLA Christian Heydemann, Gary
Warren SAIC
Partial support by DARPA GloMo Program
2
Warfighters Information Network (WIN)
How does the network perform as it is scaled to
100,000 heterogeneous devices?
OSPF, FishEye, or WRP routing protocol ?
3
State of the Art
For large models, choose efficiency or accuracy
(but not both!)

• Efficiency typically achieved via abstract
  models

• Accuracy Requires very long execution times.

4
Objectives and Approach

• Efficiency and accuracy via parallel and
  hw-in-the-loop model execution on diverse
  parallel architectures (PARSEC)

• Modular and composable library of parallelized
  models with standard APIs for end-end models
  (GloMoSim)

5
GloMoSim Library

• Model implemented using PARSEC simulation
  environment
• Modular, layered stack design
• Standard API for composition of protocols across
  different layers
• Built-in measurements on each layer
• Comparative performance evaluation of alternative
  protocols at each layer
• Scalability via support for parallel execution

6
Models available in GloMoSim
Glomosim Standalone

• Application Replicated file system, Tcplib, web
  caching, NetMeeting, WebPhone, synthetic traffic
  generators
• Transport TCP(FreeBSD), NS TCP (Tahoe), UDP,
  RTP, RSVP
• Multicasting ODMRP, CAMP, AMRIS, AMRoute, AST,
  DVMRP, FGMP
• Routing Distributed Bellman-Ford, OSPF,
  Flooding, Fisheye, DSR, WRP, LAR, DREAM, MMWN
• Link layer adaptive protocols for IP speech (MS)
• MAC CSMA, FAMA, MACA, IEEE 802.11, MACA-W,
• Radio DS SS with and without capture,
  trace-based Markovian model of fading channel
  radio (MS)
• Propagation analytical(free space, Rayleigh,
  Ricean), SIRCIM, TIREM
• Mobility random waypoint, group mobility

7
Impact of Detailed Channel Models

• Detailed channel models yield results that
  display different performance and different
  phenomena!
• Computation time for statistical models is
  10-100x that of free space models

8
Performance of detailed models

• Detailed channel models are significantly more
  accurate but also computationally expensive
• Significant Speedup for such models with parallel
  execution

9
Scalability 2
TCP Performance with Mobility
Routing Bellman Ford MAC 802.11 Mobility 100
kmph traffic CBR 100kbps tcp sessions
nodes/5
10
Complete Analysis Capability

• Efficient, high-fidelity models are only part of
  the solution

Simulation
11
SEAMLSS Environment
Mission Description
Scenario Component Selection
Radio Network Application Models
Simulation
Task Level Metrics
Automated Scenario Synthesis
Programmable Realistic Communications
CONOPS
OPNET
OPNET
OPNET
OPNET
OPNET
OPNET
RTI Federation Manager
SAF (optional)
OPNET
PARSEC
Thread Manager
SNMP Gateway
RF Propagation Model
Platoon Leader
Squad member
Squad Leader
FSM
FSP
FSQ
HLA/RTI
12
Communications CONOPS

• Thread A series of communications completing an
  operational task. Examples
• Call for fire Conduct the attack
• Situation Awareness Cross a danger zone
• Data types voice, video, data, multi-media
• Other aspects
• Service type Priority
• Encryption QoS
• Unicast, multi-cast, broadcast
• Key CONOPS metric
• Thread completion time ltgt Task Completion
  Time

13
Existing Scenarios

• Building Clearing by Team
• Halt Attack with Remote Fire Support
• Take and Defend Hill

14
Building Clearing Scenario

• Vignette Mission
• Clear 4 sections of building, upstairs and down,
  containing 2 enemy soldiers in unknown location.
• Features
• 15 minute SUO vignette
• 10 radio carrying entities -- Team TOC
• Field executable at Ft. Benning
• Dismounted infantry, highly mobile
• High fidelity terrain indoors and outdoors
• Variations
• Baseline
• 3 variations planned for different communications
  CONOPs with different bandwidth requirements. (In
  process)

Intended for validation of models against field
experiments
15
Building Clearing ScenarioInfantry Paths Over
Terrain
Building Clearing by team, Using Ft. Benning
Terrain/Features
Second Floor
Approach
First Floor
Team outside approaching building
ModSAF model complete including AutoCAD files for
Ft. Benning building
Team in room to room search
16
SEAMLSS Presented Load for Baseline
Communications CONOPs

• Integrated presented load vs. time for video,
  data and voice threads
• Scenario McKenna building clearing

17
Results From PARSECIn SEAMLSS (1)

• Scenario
• McKenna building clearing with baseline load.
• Protocols
• UDP WRP FAMA WaveLAN radio
• UDP 802.11
• Propagation
• Free space

18
Results From PARSECIn SEAMLSS (2)

• Building Clearing Scenario
• Protocols
• Fixed Protocol Stack
• TCP, IEEE 802.11
• Varied Routing Protocols
• Bellman-Ford
• OSPF
• Wireless Routing Protocol (WRP)
• Propagation
• Free space

19
Protocol Comparison

• Building Clearing Scenario
• Wireless network performance is particularly
  impacted by lower layers
• Three protocols at the MAC layer were studied
  CSMA, FAMA, IEEE 802.11

20
GloMoSim and ModSAF Co-simulation

• ModSAF (Modular Semi-Automated Forces) models
  munitions, group movement behavior
• ModSAF supports HLA through a DIS/HLA gateway
• GloMoSim, being written in PARSEC, supports HLA
  extensions
• HLA Interactions between MODSAF GloMoSim
• ModSAF sends unit positions through HLA
• GloMoSim receives position updates, computes
  signal transmission based on new positions

21
Execution Constraints

• ModSAF position updates are real-time, while
  GloMoSim/PARSEC is a DES
• an intermediate PARSEC federate was created
  between the gateway and GloMoSim

DIS-HLA Gateway RO
Intermediate Federate (IF) Time Regulated
MODSAF Real time
GloMoSim Time Constrained
22
ModSAF, DIS/HLA, Intermediate Federate, GloMoSim
DIS-HLA
IF
MODSAF
GloMoSim
RTI
23
Technology Transfer

• Parsec and GloMoSim have been integrated into
  SEAM LSS and SEAM LSS-Lite
• PARSEC downloads close to 1000
• PARSEC port to MacOS by MIT Lincoln Lab
• Over 150 GloMoSim downloads
• Second Parsec workshop to be held in November 1999

24
Parsec Downloads by Region
As of 4/21/99

• United States Fifty States
• East Asia Japan, Korea, Singapore, Taiwan,
  Malaysia, Thailand, Indonesia
• Europe Germany, Italy, Netherlands, Sweden, UK,
  Greece, Finland, France, Czechoslovakia,
  Switzerland, Spain, Belgium, Poland
• Latin America Brazil, Columbia, Venezuela,
  Mexico
• Middle East Israel, Turkey, Egypt, Iran
• Other Canada, Australia, South Africa

25
Future Work

• Demonstrate detailed simulation models with
  100,000 network nodes
• PARSEC-OPNET Cosimulation Capability
• HLA Interface for parallel PARSEC
• Software/Hardware in the loop models