NITRD/LSN Workshop On Complex Engineered Networks
Post on 24-Feb-2016
38 Views
Preview:
DESCRIPTION
Transcript
NITRD/LSN Workshop OnComplex Engineered Networks
September 20-21, 2012Washington DC
Sponsored byAFOSR
NSFDOE
Example A: Detection of Low-level Radiation Sources• Sources of low-level radiation
– small amounts of radioactive material
Overall Task: • Detect, localize and track
sources based on sensor measurements
Different versions of this task are of importance to:
• Department of Energy• Domestic Nuclear Detection
Office• OthersSeveral underlying foundational areasrelated to detection networks are open
Difficulty of Detecting Low-level Radiation SourcesThe radiation levels are only slightly above the background
levels and may appear to be “normal” background variations
• Varied Background: Depends on local natural and man-made sources and may vary from area to area
• Probabilistic Measurements: Radiation measurements are inherently random due to underlying physical process – gamma radiation measurements follow Poisson Process
Several solutions are based on thresholding sensor measurementsWell-Studied Problem: for decades using single sensors: analytical
and experimental• networks offer “newer” solutions but also raise complex design,
analysis, operation and deployment issues
Analysis Area:Quantification how much “better” a network of sensors performs
compared to single-sensor, co-located and collective detectors
Individual, Lo-Located, Collective and Network detection
: , ; 1, 2, , ;L H ii i N n n L
ˆ:;
Sn nNww
L
Individual Sensor Detection with thresholds
Co-located Sensors: Detection with threshold
Network Detection with Localization: localize the source first and detect it
: , ;L H
n n L
Collective Detection with threshold
: , ;L H
n nN wwL
Relative Performance: Individual, Lo-Located, Collective and Network detectionfor SPRT based detection methods
: ,ˆ:
ˆ ˆ : , : ,: :
: , : ,
L H
L H L H
L H L H
in Mi S
n M n iw S S
M nw
LLL L L L
L L
network with
localizationsensor with
threshold
Mn
superiority due to localization
superiority due to more measurements
: ,L Hi L
ˆ:w SL
sensor detection with threshold
co-located sensors with thresholdnetwork with
localization
: ,L H L
1, , ,N
By combining currently available analytical results
- theory leads to new detection method
Example B: Long-Haul NetworksLong-haul sensor networks: • sensors distributed across the globe and/or in space• different from well-studied “smaller” sensor networks
Application Areas:• monitoring greenhouse gas emissions using satellite, airborne,
ground and sea sensors - DOE• processing global cyber events using cyber sensors over
Internet • space exploration using network of telescopes on different
continents• target detection and tracking for air and missile defense - DOD
Response time requirements:• seconds: detecting cyber attacks on critical infrastructures -
DOD• years: detecting global trends in greenhouse gas emissions -
DOE
6
commandand
control
STcom
com
com
Sensor State Estimators
Sensor/Estimator:• Sensors/estimators generate state estimates of
dynamic targets and send them over long-haul links to fusion center
Dominant Factors:• State estimates have errors with biases • Correlations in sensor estimate errors
Cases:• Single target – single estimator stream• Multiple targets – multiple estimator streams
time-stampedstate estimates
Sensor/Estimator
phenomenon
7
Quality of fused estimates
Quality of fused estimate of sensor estimates with allocated time
8
Network:
loss, delay
Time window [t,t+W]
Correlation and fusion
Sensor estimates that reach fusion center within
used by correlation and fusion algorithms with resultant quality
A lower bound on the probability:
A Lower Bound: Guaranteed PerformanceQuality of fused state: including both network and computation effects
9
Message loss probability:• estimated based on
network parameters and communication protocols
Expected quality of correlation and fusion:• estimated based on test
measurements
Example C: UltraScience NetExperimental network research testbed:
To support advanced networking and related application technologies for large-scale projects
Currently funded by Department of Defense; by Department of Energy (2004-2007)
Features End-to-end
guaranteed bandwidth channels
Dynamic, in-advance, reservation and provisioning of fractional/full lambdas
Secure control-plane for signaling
Peering with ESnet, National Science Foundation CHEETAH, and other networks
Provides 10 Gbps dedicated connections
USN 10G Emulation
10GigE
Emulation Purpose: Continued functionality of 10G USN (de-commissioned)
Collect measurements on emulated connections Apply segmented regression to approximate USN
measurements• Emulates connection lengths not feasible on USN at much lower
cost linuxhost
linuxhost
ANUE10GigE
emulator Nexus
linuxhost
linuxhost
ANUEOC192
emulator
E30010GigEswitch
OC192
Overview of Network Simulations, Emulations and Realizations
Implementation strengths limitations
Analytical modeling
mathematical models and software
rigorous analysis and design
challenge to achieve right abstraction
SimulationsOPNET and OMNINET
software on workstations
broad what-if capability
limited reflection of networks
ANUE emulations
laboratory hardware
closer to actual network
high-cost; non-mobile
Fiber loop or USN connections
laboratory hardware, fiber spools
closest to actual network
highest cost; non-mobile
Approach: 1. Collect simulation or emulation measurement for2. Apply differential regression to obtain the estimate
Differential Regression Method for Cross-Calibration
Basic Question: Predict performance on connection length not realizable on USNExample: IB-RDMA or HTCP throughput on 900 mile connection
( )SM d Measurements on OPNET simulated path of distance d
Measurements on USN path distance
d
d
( )EM d Measurements on Anue emulated path of distance d( )U iM d id
(.)AM Regression of measurements on , ,A S E UMeasurement Regression: for
Differential Regression: for
, (.) (.) (.)A B A BM M M
, , , , ,A S E U B S E U
,ˆ ( ) ( ) ( )U C C UM d M d M d
,C S E
simulated/emulatedmeasurements
point regressionestimate
Wide-Area Infiniband Throughput:18 Different Configurations: physical and
emulatedphysical connections
emulated connectionsEmulations provide good approximation at very low cost
• Measurements collected ANUE-emulated USN connections used for interpolation/extrapolation – compared with emulated connections
• Interpolation/extrapolation:– Apply differential regression to obtain USN predictions– Interpolation: 100 and 150ms
• Not feasible on USN– in-between lengths
– Extrapolation: 200 ms• Not feasible on USN
– too long• Interpolation and extrapolation:
– For 10Gbps ANUE network emulators can provide measurements for connection lengths not feasible (too long or in-between) on USN
• Enable us to continue 10Gbps testing after 10Gbps USN de-commissioning
Analysis of iperf and disk transfer measurements
top related