CS321: Information Processing for Sensor...
Transcript of CS321: Information Processing for Sensor...
![Page 1: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/1.jpg)
1
CS321: Information Processing CS321: Information Processing for Sensor Networksfor Sensor Networks
Leonidas GuibasComputer Science Dept.Stanford University
Sensing Networking
Computation
![Page 2: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/2.jpg)
2
Introduction
Distributed algorithmsNetworkingDatabasesSoftware radiosSoftware design
Low-power processorsSignal processingWireless communicationInformation theoryEstimation theory
CS EE
![Page 3: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/3.jpg)
3
Rockwell HiDRA
Environmental sensingTraffic, habitats, pollution, hazards, security
Industrial sensingMachine monitoring and diagnostics (IC fab)Power/telecom grid monitoring
Human-centered computing
Smart, human-aware spaces and environments
Berkeley/Crossbow Motes
Untethered micro sensors will go anywhere and measure anything -- traffic flow, water level, number of people walking by, temperature. This is developing into something like a nervous system for the earth. --Horst Stormer in Business Week, 8/23-30, 1999.
Smart Sensors and Sensor Networks
UCLA WINS
Sensing Networking
Computation
![Page 4: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/4.jpg)
4
Wireless Sensor Networks
Distributed systems consisting of small, untethered, low-power nodes capable of sensing, processing, and wireless communication
small
large
RFID
PDA
Sensoria Node
MS Spot Watch
![Page 5: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/5.jpg)
5
Monitoring the World
Monitoring the environment and other spacesMonitoring objectsMonitoring interactions between objects, or between objects and their environment
![Page 6: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/6.jpg)
6
Petrel Nesting Behavior at Great Duck Island
![Page 7: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/7.jpg)
7
Wireless Sensor Network Deployment
Advantages:sensors can be close to signal sources, yielding high SNRcan monitor phenomena widely distributed across space and timedistributed architecture provides for scalable, robust and self-repairing systemssignificant savings on cabling, etc are possible
British Columbia winerywith networked temperaturesensors
Other data collection and monitoring: temperaturein data centers (HP), oil tanker vibrations (Exxon),soil contaminants, etc.
![Page 8: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/8.jpg)
8
Integration with Current Networks
Access to unfiltered information, highly localized in time and space
![Page 9: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/9.jpg)
9
More Demanding Sensor Network Applications
Beyond simple data collection and aggregation
acting on the worldsimultaneous tracking of multiple objectsdistributed, wide-area phenomenadistributed attention: focus and context
Network must adapt to highly dynamic foci of activitySensing and communication tasks must be allocatedResources must be apportioned between detection, tracking, etc.
![Page 10: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/10.jpg)
10
Sensor Network Hardware
![Page 11: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/11.jpg)
11
Wireless Sensor Trends
Riding on Moore’s law, smart sensors get:
Of 9.6 billion µP’s to be shipped in 2005, 98% will be embedded processors!
Sensoria WINSNG 2.0CPU: 300 MIPS 1.1 GFLOP FPU32MB Flash32MB RAMSensors: external
More powerful
Crossbow Mica2dot mote4 MIPS CPU (integer only)8KB Flash512B RAMSensors: on board stack(accel, light , microphone)
Inexpensive & simple
Smart dust (in progress)CPU, Memory: TBD (LESS!)Sensors: integrated
Supercheap & tinyEasy to use
HP iPAQ w/802.11CPU: 240 MIPS32MB Flash64MB RAMBoth integrated and off-board sensors
![Page 12: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/12.jpg)
12
Most Popular: Crossbow Motes
51-pin MICA2 / GPIO Connector
Buzzer
Light & Temperature Sensor
Microphone
Chipcon CC2420 802.15.4 Radio
Atmel ATMega128L
(under)
![Page 13: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/13.jpg)
13
Crossbow Stargate - Top View
Ethernet RJ-45 USB
Serial Port RS-232
PCMCIA SLOT
![Page 14: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/14.jpg)
14
Crossbow Stargate - Bottom View
Compact Flash Slot
SA 1111 StrongArm I/O Chip
Intel PXA255 XscaleProcessor
51-pin MICA2 / GPIO Connector
![Page 15: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/15.jpg)
15
SpecificationsStargate
Embedded Linux OS400 Mhz Intel Xscale64 MB SDRAM32 MB FLASHMany different interfaces
RS-232, Ethernet, USB,…
MicaZ Mote• TinyOS• 16 Mhz Atmel ATMega128L• 128 kB Program FLASH• 512 kB Serial FLASH• Current Draw
• 8 mA – Active Mode• <15 uA – Sleep Mode
• Chipcon CC2420 802.15.4 Radio• 250 kbps• 26 Channels – 2.4 Ghz• Current Draw – 15 mA
www.xbow.comhttp://computer.howstuffworks.com/mote4.htm
![Page 16: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/16.jpg)
16
Power Breakdown …
Panasonic CR2354560 mAh
This means– Lithium Battery runs for 35
hours at peak load and years at minimum load, a three orders of magnitude difference!
003 mAEE-Prom
000
4.5 mA (RX)2 mA
Idle
0200 �ATemperature0200 �APhoto Diode04 mALED’s
5 �A7 mA (TX)Radio5 �A5 mACPU
SleepActive
Rene motes data, Jason Hill
Computation/communication ratio:
• Rene motes:
• Comm: (7mA*3V/10e3)*8=16.8�J per 8bit
• Comp: 5mA*3V/4e6=3.8 nJ per instruction
• Ratio: 4,400 instructions/hop
• Sensoria nodes:
• Comm: (100mW/56e3)*32=58�J per 32bit
• Comp: 750mW/1.1e9=0.7nJ per instruction
• Ratio: 82,000 instructions/hop
![Page 17: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/17.jpg)
17
ArchitecturalChallenges
![Page 18: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/18.jpg)
18
Sensor Network ChallengesPower management
communication 1000s of times more expensive than computationload balancingcoordinated sleeping schedulescorrelated sensor data
In-network processingdata aggregationovercounting of evidence
Difficult calibrationlocalizationtime-synchronization
Constant variabilitynetworkingsensing
![Page 19: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/19.jpg)
19
Distributed Sensor Communication: Multi-Hop RF Advantage
( ) 1
( )
( )send Nr receive
send r receive
P Nr PN
N P N r P
αα
α−= =
⋅ ⋅
Power advantage:
send receiveP r Pα∝Or equivalently,
, :3 5sendreceive
PP
rα α∝ −
RF power attenuation near ground:
receiveP
( )send rP
( )send NrP
r
Nr
![Page 20: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/20.jpg)
20
Semantic Routing and Networking
We want to address spatial locations or information, not individual nodesContent and address in a message get intermixed How do we help information providers and clients find each other? Directed diffusion
Geo-routing
![Page 21: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/21.jpg)
21
In-Network Processing
Information aggregation can happen on the way to a destinationNeed to balance quality of paths with quality of information collectedAre there “application-independent” paradigms of information aggregation?
Temperature aggregation
![Page 22: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/22.jpg)
22
Networking Sensor Networks
Network support for a small number of collaborative tasks.Data-centric, (as opposed to a node-centric) view of the world.Monitoring processes may migrate from node to node, as the phenomena of interest move or evolve.Communication flow and structure is dictated by the geography of signal landscapes and the overall network task.
![Page 23: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/23.jpg)
23
Distributed Sensor Networks: Detection and SNR Advantage
2source
receive
PP
r∝Acoustic power received at distance r:
10log 10log 20logr source noiseSNR P P r= − −
Signal-noise ratio (SNR):
20log 10logr rN
rSNR SNR N
rN
− = =
Increasing the sensor density by a factor of N gives a SNR advantage of:
Sensors have a finite sensing range. A denser sensor field improves the odds of detecting a target within the range. Once inside the range, further increasing sensor density by N improves the SNR by 10logN db (in 2D). Consider the acoustic sensing case:
![Page 24: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/24.jpg)
24
Collaborative EstimationStructuring communication is very important:
In a setting where each node wishes to communicate some data to another node at random, interference hinders scaling:
the per node throughput scales as (Gupta & Kumar ‘99)
Effectively each node is using all of its energy to routemessages for other nodes.
In a sensor network, however, because data fromnearby sensors are highly correlated and moreintelligent information dissemination strategies arepossible.
1N
![Page 25: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/25.jpg)
25
Power-Aware Sensing and Communication
Variable power systemsLet most sensors sleep most of the time; paging channelsExploit correlation in readings between nearby sensorsLoad-balance, to avoid depleting critical nodes
Wireless communication with neighboring nodes
In-node processing
![Page 26: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/26.jpg)
26
Sensor Tasking and Control
Decide which sensors should sense and communicate, according to the high-level task – a non-trivial algorithmic problemDirect sensing of relations relevant to the task – do not estimate full world state
d ahead-of e
c ahead-of d
b ahead-of c
a ahead-of b
![Page 27: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/27.jpg)
27
Enable Data-Base Like Operations
Data only available right after sensing operationDense data streams must be sampled, or otherwise summarizedMust deal with distributed information storage –“where is the data?” Field isolines
![Page 28: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/28.jpg)
28
Self-Configuration for Ad-Hoc Deployment
Network size makes it impossible to configure each node individuallyEnvironmental changes may require frequent re-calibrationNetwork must recover after node failures
Iterative localization
![Page 29: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/29.jpg)
29
Structure Discovery
A sensor network is a novel type of computing device -- a sensor computerOne of its first tasks is to discover its own structure and establish
information highwayssensor collaboration groups
as well as adapt to its signal landscape
![Page 30: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/30.jpg)
30
New System Architectures
Resource constraints require close coupling between the networking application layersCan we define application-independent programming abstractions for sensor networks?
In-network: application processing, data aggregation, query processing
Adaptive topology, geo-routing
MAC, time and location services
Phy: comm, sensing, actuation, SP
User queries, external databases
Data dissemination, storage, caching
A sensor net stack?
![Page 31: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/31.jpg)
31
Various Issues
Integration of sensors with widely different modalities
High data-rate sensors (cameras, laser scanners)
Sensor mobilityActuation
Distributed robotics
![Page 32: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/32.jpg)
32
What Defines Sensor Networks?
Multi-hop communicationMany nodes act as routersMultiple paths exist and must be considered
Bandwidth limitationsVolume of data sensed exceeds to capacity of the network to transport
Power limitations(At least some) nodes operate untethered and energy conservation must be considered in all of sensing, processing, and communication
A cooperative systemAll nodes serve one, or a small number of tasks
![Page 33: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/33.jpg)
33
Sensor Network Researchpower awarenesssensor tasking and controlformation of sensor collaboration groupsin-network, distributed processingnode management, service establishment, software layerscoping with noise and uncertainty in the environment
Estimate fullworld-state
Sense Answer query,make decision
A key algorithmic problem is how to sense and aggregate only the portions of the world-state relevant to the task at hand, in a lightweight, energy-efficient manner.
![Page 34: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/34.jpg)
34
CS321: The Course
![Page 35: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/35.jpg)
35
Course Outline, I
Networking sensor networks (topology discovery, MAC layer issues, routing, energy considerations)Information brokerage (directed diffusion, distributed hash tables)Infrastructure establishment (time-synch, localization)Sensor network hardware (Motes/Stargates and other common node types, power issues)
![Page 36: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/36.jpg)
36
Course Outline, II
Sensor network software (TinyOS, programming abstractions)Probabilistic state estimation with sensor data (Kalman and particle filters)Sensor tasking and control (value of information, collaborative signal processing)Distributed data processing and storage (in-network data aggregation, range searching)Applications (incl. camera networks)
![Page 37: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/37.jpg)
37
Course TextBook
Wireless Sensor Networks: An Information Processing Approach
Geng Zhao and Leonidas Guibas
Morgan-Kaufmann 2004
![Page 38: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/38.jpg)
38
Course Format
Lectures by the instructor, plus a few guest lecturersStudent presentations of current research papersA course project is required. Can be
an implementation on actual mote hardwarean extensive simulationa more theoretical investigation
Alternate meeting for Monday’s class
![Page 39: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/39.jpg)
39
Course Personnel
Instructor:Leonidas J. Guibas
TA:Qing Fang
Course web page:
http://graphics.stanford.edu/courses/cs321-05-fallhttp://www.stanford.edu/class/cs321
![Page 40: CS321: Information Processing for Sensor Networksgraphics.stanford.edu/courses/cs321-05-fall/Lectures/05cs321-sep26.pdf · CS321: Information Processing for Sensor Networks Leonidas](https://reader031.fdocuments.net/reader031/viewer/2022022716/5c17410609d3f228458b8ad7/html5/thumbnails/40.jpg)
40
The End