Transcript PP slides

Network Design
1 May 2008
Dr. Charles Graff,
US Army RDECOM
CERDEC-STCD
Ft. Monmouth NJ 07703
Briefing Outline
• Background on Networks
• Network Design Issues
• STCD Network Design Program
3/26/201610/21/04
CERDEC-021.2
Distribution Statement
• This briefing has been previously cleared for public
release.
• The comments recommendations and conclusions are
solely those of the author, and not of the Army or DOD.
3/26/201610/21/04
CERDEC-021.3
Background on Networks
Mobile Ad Hoc Networks (MANETs) vs
Commercial Networks
• Mobile Ad Hoc Networks (MANETs) do not have a fixed infrastructure
or backbone
• MANET connectivity is determined by node location, mobility, and RF
propagation characteristics.
• Commercial Networks ( Digital Cellular ) have fixed infrastructure
backbone of Base Stations and Cell Towers that are interconnected
thru high bandwidth Fiber Optic cable.
• In Digital Cellular, only the “last hop” is wireless.
• Other Commercial wireless networks ( IEEE 802.xx ) are usually
static in nature.
• The MANET model of networks, with extensions for multi-tier
operation, is most applicable to Army Networks such as FCS and
WIN-T.
3/26/201610/21/04
CERDEC-021.5
MANET Networks
• User Traffic and RF connectivity are in fact probabilistic and
stochastic quantities
• User node mobility is also a stochastic quantity
• Node interconnection, as determined by RF propagation, leads to a
time dependent, stochastic topology graph
- Classic Graph Theory has limited application for MANETS due to the
inability of represent link to link coupling and node mobility explicitly.
• Hence the mathematics of networks must include both a stochastic
nature as well as a combinatoric nature
• In WIRELESS MANETS, both the Network and the Traffic loads
are stochastic in nature
- In WIRED Networks typically only the Traffic is stochastic
• Hence MANET Network Design is a HARD problem!!
3/26/201610/21/04
CERDEC-021.6
Network Analysis vs. Network Design (Synthesis)
• The Analysis Problem:
- Given a network solution perform an performance analysis
typically for
» User Requirements for throughput, delay, reliability
» Network survivability
» Network connectivity
• The Design Problem:
- Design is the Synthesis Problem
- Given what you want ( as given requirements), how do you go
about creating the network solution that meets (or exceeds the
requirements?
- This is a creative and inventive process
3/26/201610/21/04
CERDEC-021.7
Network Design Issues
Specific Network Design Issues
• Design metrics for MANETs
- Connectivity ( in mobile ad hoc environment )
- ETE User Requirements
- Survivability ( in mobile ad hoc environment )
- “Optimality” or goodness of design to be used in comparisons of
different designs
• RF Connectivity representation ( including mobility )
• Steady state vs transients in network operation
• Closed loop vs open loop for adaptive/dynamic algorithms
• Scalability ( to arbitrary network sizes )
• Multi-tiered Operation
- Widely differing RF characteristics link highly dispersed users
• Design solution should be “insensitive” to traffic loads, operational
scenarios
- Typically requirements are uncertain and/or subject to change
3/26/201610/21/04
CERDEC-021.9
Specific Network Design Issues (Concluded)
• Security
• “Optimality” in the face of uncertain requirements,
scenarios, and RF environments
• RF Waveform Design for Jamming Environment
• “Validation” issues
- Does the design meet/exceed requirements
- Typically requires extensive testing
• “Verification” Issues
- Does the design have “good properties” that are
common to all designs
• COST, COST, COST
3/26/201610/21/04
CERDEC-021.10
Network Analysis / Design Techniques
• Analytic Modeling
- Define/develop mathematical equations for network behavior –
connectivity, capacity, thruput/goodput, delay, survivability, etc
- Very difficult to get good closed form solutions
- A 6.1 and academic focus
• Simulation
- Develop Discrete Time Event Driven simulation for the Network
- Execute simulation over range of traffic loads, topologies, mobility
patterns, failure/destruction patterns
- Requires a large number of runs to get solutions
- OPNET, Qualnet, etc
3/26/201610/21/04
CERDEC-021.11
Network Analysis/Design Techniques ( Concluded )
• Prototyping
- Build a solution, make measurements on it.
- Collect a large amount of data and then use data driven modeling
techniques to define necessary relationships
- Rutgers WIN-LAB ORBIT is good example
• Emulation
- Build a scaled or abstracted version of the network solution
- make measurements and develop data driven relationships
3/26/201610/21/04
CERDEC-021.12
Mathematics Potentially Applicable to Network
Design and Analysis
• Graph Theory
- Topology representation and analysis – but not sufficient for MANET
• Closed Loop control theory
- Assumes a “network operating point” that is to be maintained
- Need to add delay/ loss model to control model
• Stochastic Processes
- “Random” or statistical nature of traffic load, node mobility, and RF channel,
Discrete Time Markov Processes, Fractals, Jackson Networks. BCMP
Networks
• Discrete Event Transition Models
- Finite State machines, Petri-Nets, etc
- To represent control interaction with system state
• Optimization Theory as applied to Network Design
- Primal/Dual, hill climbing, annealing, Genetic Algorithms, Robust
Optimization, Stochastic Optimization
• VERY MESSY INDEED
3/26/201610/21/04
CERDEC-021.13
Overall MANET Network Design Elements
• The protocol/stack design
- individual protocols
- cross layer issues
- protocol parameters/configuration
• The “node” design
- Radio hardware design
- Antenna Modifications
- Waveform Modifications
• The Recommended Network Architecture
- Relay function for multi-hop networks
- Addition relay (like UAV or UGV) capability may be needed to
achieve network goals of connectivity, capacity, and survivability
- Domain sizing and organization
- Interconnection points
3/26/201610/21/04
CERDEC-021.14
Cross Layer Design Issues (Goldsmith, et. al.)
• Multiple Antennas
• Coding/Modulation
• Power Control
• Adaptive Link Techniques for Power Control, Scheduling,
and link selection
• Neighbor selection / maintenance
• Delay/energy constrained routing
3/26/201610/21/04
CERDEC-021.15
Network Layer Design Issues
• Network Initialization Time (Cold Start)
• Node Join Time To Existing Network
• Group Node Join Time
• Node Leave Time From Existing Network
• Group Node Leave Time
• Network Recovery Time ( after node / link failure )
• Network Overhead ( packets/sec )
• Processing Resource Requirements ( CPU
cycles/memory)
• QOS/Data Handling
3/26/201610/21/04
CERDEC-021.16
Application Layer Issues
• ETE delay, throughput/goodput, packet loss
• Reliability/Connection Management
• QOS/priority
3/26/201610/21/04
CERDEC-021.17
STCD Network Design Program
Background on Engineering Network Design Tools
• Design Tools exist for wired backbone networks, such as
Digital Cellular and POTS.
• Some design tools exist for IEEE802.xx type networks,
but are typically limited only to connectivity determination
and node placement.
• Some design Tools exist for Satellite Networks but are
limited primarily to up/down link design.
3/26/201610/21/04
CERDEC-021.19
Why a Network Design Toolset?
• Due to network complexity, many relationships may be required
to accurately describe network behavior
• These relationships will need to be coupled and use in a
coordinated fashion to produce a complete network design
• Tools/Toolsets have be used successfully in many other areas:
- Cell phone network design
- Design of Boeing 777
- VLSI circuit design
- Layout of internal spaces on submarines
• But not ad hoc network design ( yet!)
3/26/201610/21/04
CERDEC-021.20
Network Design vs Network Planning
• Network Planning
- Used by Military user in Operational environment to plan
deployment of existing network assets to meet
operational needs in theater.
- Assumes that Network hardware/software is fixed and
designed.
• Network Design
- Used by Engineering community
- protocol/stack design, node design, and network
architecture
- provides rapid design/trade offs of design options
- validated thru detailed OPNET ( or equivalent )
simulations, prototyping, or experimentation
3/26/201610/21/04
CERDEC-021.21
Approaches to Network Design
• Analytic Approach
- Human must create, discover or invent mathematical relationships/heuristics
- Many assumptions required for mathematical tractability
- Typically, one function at a time is all that can be represented mathematically ; hence
many coupled relationships may be required to completely describe ad hoc networks
- In many areas, these analytic relationships do not exist
• Discrete Time Event Driven Simulation Approach
- The human must do the design, and the simulation does the computation; The human
does the analysis of simulation results
- Very detailed, fine grained simulations are possible
- Specific solutions to specific input sets (a big calculator)
- Many runs needed to vary parameters and get statistically valid results
- Little to no insight as to network behavior or “what effects what”
• Prototyping / experimental testing Approach
- Must have a network built first
- Instrumentation issues to get accurate measurements
- Much data can be collected --> challenging data analysis
SHOULD BE DONE TOGETHER IN CONSISTENT FASHION to achieve good network design
3/26/201610/21/04
CERDEC-021.22
Network Engineering Design Analytic Toolset (NEDAT)
Network output - What can
the network do?
Network input goals/ranges.
NEDAT INPUTS
• NUMBER OF NET NODES
• LOCATION OF NET NODES
• RANGE OF XMIT POWER
• RF REPRESENTATIONS
• RANGE OF RF LINK
RATES
• NET CONNECTIVITY
• NET END TO END
CAPACITY
NEDAT OUTPUTS
NEDAT
Development Approach
“Build a little/test a little”
• Start with small numbers
of nodes (~15)
• Increase number of nodes
(~100 , ~500, ~1000)
• Add more refined behavior
as Network Science results
become available.
• GENERIC PROTOCOLS
3/26/201610/21/04
CERDEC-021.23
• ACTUAL XMIT
POWER
• ACTUAL DATA
RATES
• NET SURVIVABILITY
• MOBILITY PATTERNS
• ADDITIONAL NODES
AT LOCATIONS
NEEDED??
6.1 RESEARCH NETWORK DESIGN
EQUATIONS/RELATIONSHIPS
• PROTOCOL STACK
REQUIRED
PARAMETERS
Network Engineering Design Analytic Toolset (NEDAT)
Detailed OPNET Simulation used to “validate” Network Design
Produced by NEDAT
Scenario/Traffic Load
Connectivity,
Capacity,
Survivability
High Level Node
Representation
Design Equations
Heuristics From
Network Science
NEDAT
Discrete Time
Event driven
OPNET
Simulation
Network
Design
No
Goals Met?
Ideas for Refinement
Yes
Recommend Lab/Field
Experimentation
3/26/201610/21/04
CERDEC-021.24
What is Cognitive Networking (CN)?
• Cognitive Networks (CN) are characterized by advanced hardware and
software that
- Interacts proactively with the environment (RF, traffic load, mobility, mission profile,
etc)
- Uses learning, knowledge representation, estimation, predictive and optimization
techniques for (near) real time network control (i.e. protocols), and network and
spectrum management
- Provides enhanced performance (user thruput, delay, loss, survivability) and spectral
efficiency over “conventional” techniques
» Conventional techniques use fixed algorithms with variable parameters, while
Cognitive Networks use variable algorithms with variable parameters in stable,
non-oscillatory, adaptive fashion
• (MILITARY) Cognitive Networking is the application of the above
techniques to the MANET Technology
3/26/201610/21/04
CERDEC-021.25
Cognitive Networking vs “Traditional” Networking
• Cognitive Networking will solve the well known MANET networking
problems in a “different” and better way when compared to “traditional”
approaches
- Interact, learn, and respond to networking environment
- Algorithms change their computation logic (based on the environment) as well as
usual algorithm parameters
- Provide less overhead thru the use of advanced techniques (i.e. compute, don’t
communicate) such as predictive, estimation, local reasoning, etc.
• The Networking problems to be addressed remain the same
- Network Ops, Control/Management
- Protocols / Cross layer stack design
- Security
- Multi-tiered Network Transport
3/26/201610/21/04
CERDEC-021.26
Why Cognitive Networking Design (CND) is
hard?
• Mobile Ad Hoc Networking design is hard in general; the addition of
cognitive capabilities increase the design space/options
• Many options and alternatives need to be examined and explored
• Traditional simulation (i.e. OPNET) approaches have limitations
regarding model development, fidelity, scalability and are actually
analysis and not design tools
• Many of learning, estimation, reasoning, optimization and other
techniques have not been applied to the highly mobile, large scale,
distributed, dynamic ad hoc networking for the hostile battlefield
environment.
3/26/201610/21/04
CERDEC-021.27
Why a CN Design Tool?
• MANETs in general are hard to design due to the large design space
• Using Cognitive Networking technology only increases the design space
options and adds complexity to the process with the expectation of
enhanced performance and efficiency of operation.
• Using a modeling approach to present both cognitive and non-cognitive
functions in an engineering tool environment will allow rapid and effective
exploration of a large number of design options and alternatives.
• Using a modeling approach, critical issues such as scalability,
performance, and behaviors may be explored and investigated at minimal
cost without committing to large amounts of physical hardware or
expensive testing
3/26/201610/21/04
CERDEC-021.28
Cognitive Network Design (CND) Considerations
• Design metrics for MANETs
- Connectivity ( in mobile ad hoc environment )
- End-to-end User Requirements
- Survivability ( in mobile ad hoc environment )
- “Optimality” or goodness of design to be used in comparisons of different designs
• Knowledge oriented representation of RF connectivity (including
mobility), network operation/behaviors
• Effectiveness of learning/prediction techniques in dynamic environment
• Steady state vs. transients in network operation (stability issues)
• Design solution should be “insensitive” to traffic loads, operational
scenarios/requirements and environments
- Typically these are uncertain and/or subject to change
10/21/04 CERDEC-021.29
3/26/201610/21/04
CN Design Tool Software Functional Architecture
Prediction/Estimation
Learning
Functional Model
Repository
Analytic Tool-box
???
Input
Module
Design Manager
DES
Adaptor
External
DES
3/26/201610/21/04
CERDEC-021.30
Knowledge
Database
Output
Module
The Payoff of Cognitive Networking
 The ability to say, with a high level of confidence that the “network”
will work in the military dynamic environment
 A potential to perform cost-performance trade offs for various
cognitive network design through analysis for large military ad-hoc
networks
 A new capability to “optimize the design” for performance
functionality, and capabilities of mobile ad-hoc networks
 A better understanding of complex network behaviors
New capability to optimize engineering design of large, multi-tiered,
highly mobile, ad hoc networks
10/21/04 CERDEC-021.31
3/26/201610/21/04