Wireless Sensor Networks and Laboratories Polly Huang EE NTU

Communication Protocols Diffusion Routing Magnetic Diffusion Cross-Layer Performance Analysis

Directed Diffusion largely based on slides from Chalermek Intanagonwiwat & Deborah Estrin

In Short A data dissemination mechanism fitting into the data-centric communication paradigm for sensor networks

Sensor network, what? Sensor Networks Common Features Challenges Approach Why not IP based solution?

Sensors Devices to sense the situation about physical objects or environments The situations –Location, motion, visual, sound, vital signs, temperature, brightness, etc The sensors –Could be placed at close proximity of the sensing target –Could be tagged physically on to the sensing target

Sensor Networks Or another One way

Applications Scientific: eco-physiology, biocomplexity mapping Infrastructure: contaminant flow monitoring (and modeling) Engineering: monitoring (and modeling) structures

The Real Need Specialized communication in a wild wide space –Specialized: application dependent –Wild: little or no infrastructure –Wide: expensive to build/use communication infrastructure

Applications: A Longer List Science: monitoring temperature change on a volcanic island Engineering: monitoring power use of industrial district Infrastructure: monitoring passenger traffic at MRT stations Military: tracking enemy migration in a dessert Disaster: emergency relief after Gozzila taking a short tour of Tokyo

Common Vision Embed numerous distributed devices to monitor and interact with physical world Exploit spatially and temporally dense, in situation, sensing and actuation Network these devices so that they can coordinate to perform higher-level tasks Requires robust distributed systems of hundreds or thousands of devices

Challenges Tight coupling to the physical world and embedded in unattended systems –Different from traditional Internet, PDA, Mobility applications that interface primarily and directly with human users –But solutions might be applicable to the Internet, PDA, Mobility applications as well Untethered, small form-factor, nodes present stringent energy constraints –Living with small, finite, energy source is different from traditional fixed but reusable resources such as BW, CPU, Storage Communications is primary consumer of energy in this environment –R 4 drop off dictates exploiting localized communication and in-network processing whenever possible

Energy the Bottleneck Resource Communication VS Computation Cost [Pottie 2000] –E α R 4 –10 m: 5000 ops/transmitted bit –100 m: 50,000,000 ops/transmitted bit Avoid communication over long distances Cannot assume global knowledge, cannot pre- configure networks –Achieve desired global behavior through localized interactions –Empirically adapt to observed environment Can leverage data processing/aggregation inside the networkCan leverage data processing/aggregation inside the network

In-Network Processing Sensor technology is advancing steadily Situations detected by the sensors can be surprisingly rich For example, all these at once –Detecting a speech –Inferring the location and identity of the speaker These information can be used to facilitate efficient dissemination of the recorded speech –Suppressing speech coming from the same speaker –Forwarding towards the likely listeners

New Design Themes Long-lived systems that can be untethered and unattended –Energy efficient communication –Self configuring systems that can be deployed ad hoc

Approach Leverage data processing inside the network –Exploit computation near data to reduce communication Achieve desired global behavior with adaptive localized algorithms (i.e., do not rely on global interaction or information) –Dynamic, messy (hard to model), environments preclude pre-configured behavior –Can ’ t afford to extract dynamic state information needed for centralized control or even Internet-style distributed control

Why can ’ t we simply adapt Internet protocols and “ end to end ” architecture? Internet routes data using IP Addresses in Packets and Lookup tables in routers –Humans get data by “ naming data ” to a search engine –Many levels of indirection between name and IP address –Works well for the Internet, and for support of Person-to- Person communication Embedded, energy-constrained (un-tethered, small- form-factor), unattended systems can ’ t tolerate communication overhead of indirection

Therefore, Directed Diffusion Features Operations Evaluations

Directed Diffusion Paradigm Data-centric communication Supported with distributed algorithms using localized interactions Application-specific in-network processing

IP Communication Organize system based on named nodes Per-node forwarding state Senders need to push data to the node address of sink Bob Alice To Bob My name is Alice. I am a 19-yr old girl… Chris I am Bob Bob there I am Bob Bob there I am Bob To Bob My name is Alice. I am a 19-yr old girl… To Bob My name is Alice. I am a 19-yr old girl…

Data-Centric Communication Organize system based on named data Per-data diffusion state Sinks need to be specific about what data they’d pull Tell me about girls Tell me about girls Girl info goes there Tell me about girls Girl info goes there Tell me about girls Here’s a 19-yr old girl…

Directed Diffusion Paradigm Data-centric communication Supported with distributed algorithms using localized interactions Application-specific in-network processing

Localized Interaction Diffuse requests/interest across network Set up gradients to guide responses/data Diffuse responses/data based on the gradients (Pretty much the same as in the IP routing) Tell me about girls Tell me about girls Girl info goes there Tell me about girls Girl info goes there Tell me about girls Here’s a 19-yr old girl…

Directed Diffusion Paradigm Data-centric communication Supported with distributed algorithms using localized interactions Application-specific in-network processing

Without In-Network Processing Data are simply passed on Tell me about girls Tell me about girls Girl info goes there Tell me about girls Girl info goes there Tell me about girls Tell me about girls Here’s a 20-yr old girl… Here’s a 19-yr old girl… Here’s a 20-yr old girl… Here’s a 19-yr old girl… Here’s a 20-yr old girl…

With In-Network Processing Data are aggregated and then passed on Girl info goes there Here’re two 19+ yr old girls… Girl info goes there Here’s a 20-yr old girl… Here’s a 19-yr old girl… Here’re two 19+ yr old girls… Here’s a 20-yr old girl… Here’s a 19-yr old girl… Here’re two 19+ yr old girls… Application-specific Aggregation Here!

Directed Diffusion Paradigm Data-centric communication Supported with distributed algorithms using localized interactions Application-specific in-network processing

Example: Remote Surveillance Interrogation: –e.g., “ Give me periodic reports about animal location in region A every t seconds ” Interrogation is propagated to sensor nodes in region A Sensor nodes in region A are tasked to collect data Data are sent back to the users every t seconds

Basic Directed Diffusion Setting up gradients Source Sink Interest = Interrogation Gradient = Who is interested

Basic Directed Diffusion Source Sink Sending data and Reinforcing the best path Low rate eventReinforcement = Increased interest

Directed Diffusion and Dynamics Recovering from node failure Source Sink Low rate event High rate event Reinforcement

Directed Diffusion and Dynamics Source Sink Stable path Low rate event High rate event

Local Behavior Choices For propagating interests –In this example, flood –More sophisticated behaviors possible: e.g. based on cached information, GPS For data transmission –Multi-path delivery with selective quality along different paths –probabilistic forwarding –single-path delivery, etc. For setting up gradients data-rate gradients are set up towards neighbors who send an interest. data-rate gradients are set up towards neighbors who send an interest. Others possible: probabilistic gradients, energy gradients, etc. For reinforcement reinforce paths, or parts thereof, based on observed delays reinforce paths, or parts thereof, based on observed delays, losses, variances etc. other variants: inhibit certain paths because resource levels are low

Initial simulation study of diffusion Key metric –Average Dissipated Energy per event delivered indicates energy efficiency and network lifetime diffusionCompare diffusion to –flooding omniscient multicast –centrally computed tree (omniscient multicast)

Diffusion Simulation Details ns-2Simulator: ns-2 Network Size: Nodes Transmission Range: 40m Constant Density: 1.95x10 -3 nodes/m 2 (9.8 nodes in radius) MAC: Modified Contention-based MAC Energy Model: Mimic a realistic sensor radio [Pottie 2000] –660 mW in transmission, 395 mW in reception, and 35 mw in idle

Diffusion Simulation Surveillance application –5 sources are randomly selected within a 70m x 70m corner in the field –5 sinks are randomly selected across the field –High data rate is 2 events/sec –Low data rate is 0.02 events/sec –Event size: 64 bytes –Interest size: 36 bytes –All sources send the same location estimate for base experiments

Sensor radio Average Dissipated Energy (Sensor radio energy model) Average Dissipated Energy (Joules/Node/Received Event) Network Size Diffusion Omniscient Multicast Flooding Diffusion can outperform flooding and even omniscient multicast. WHY ?

Impact of In-network Processing Average Dissipated Energy (Joules/Node/Received Event) Network Size Diffusion With Suppression Diffusion Without Suppression Application-level suppression allows diffusion to reduce traffic and to surpass omniscient multicast.

Impact of Negative Reinforcement Average Dissipated Energy (Joules/Node/Received Event) Network Size Diffusion With Negative Reinforcement Diffusion Without Negative Reinforcement Reducing high-rate paths in steady state is critical

Summary of Diffusion Results Under the investigated scenarios, diffusion outperformed omniscient multicast and flooding Application-level data dissemination has the potential to improve energy efficiency significantly –Duplicate suppression is only one simple example out of many possible ways. –Aggregation (next) All layers have to be carefully designed –Not only network layer but also MAC and application level

Standard Average Dissipated Energy (Standard energy model) Average Dissipated Energy (Joules/Node/Received Event) Network Size Diffusion Omniscient Multicast Flooding Standard is dominated by idle energy

Source 1 Source 2 Sink Source 1 Source 2 Sink Late Aggregation Early Aggregation Greedy Aggregation Low-latency tree might be inefficient (late aggregation) Bias path selection to increase early sharing of paths (early aggregation) Construct greedy incremental tree (GIT) –establish t shortest path for first source –connect each other source at closest point on existing tree

Mechanisms Path Establishment –Propagate energy cost with events –On-tree incremental cost message for finding closest point on existing tree –Path selection based on lowest energy cost (events and incremental cost messages) Path maintenance –Use greedy heuristic of weighted set-covering problem to compute energy cost of an outgoing aggregate Source 1 Source 2 Sink E 2 = 0 E 2 = 2 E 2 = 1 E 2 E 2 = 2 E 2 E 2 = 3 E 2 = 4 E 2 = 2 E 2 = 3 E 2 = 4 E 2 = 5 C 2 = 2 C 2 C 2 C 2 Source 1 Source 2 Sink Incremental cost message Reinforcement

Evaluation Greedy aggregation appears to outperform opportunistic aggregation only in very high-density networks opportunistic greedy

Testbed Experiments

Proof-of-Concept Experiment: Nested Queries Edge processing overwhelms power and bandwidth consumption Nested queries where low-energy sensors trigger high-energy sensors Edge Processing Nested Queries with In-network Processing

Nested Queries Used BAE-Austin ’ s signal processing –Live, Multiple-target, real-vehicle detections SITEX ’ 02 validates previous lab experiments –Reduces network traffic/Improves event delivery ISI Testbed Data: 2-level are nested queries29Palms Data nested end-to-end event delivery ratio

Questions?

Ad Hoc Network Routing

Ad Hoc Network A collection of wireless mobile nodes Dynamically forming a temporary network

Features Without the use of any existing network infrastructure or centralized administration –Infrastructure-less networking Little or no communication infrastructure Expensive or inconvenient to establish/use infrastructure –No central administration Some overlay network Some peer-to-peer networks

Ad Hoc Routing Finding a path from the source to the destination in ad hoc networks Multi-hop exchange Each host is also a router

Temporally-Ordered Routing Algorithm (TORA) Presented INFOCOM ’ 97 by Park and Carson Designed to Minimize overhead and discover routes on demand Think about it as water flowing through tubes on its way to a destination Node broadcasts a QUERY packet, recipient broadcasts an UPDATE packet Uses IMEP as transport –Reliable, in-order transmission

Route Creation Example

Magnetic Diffusion

Sensor Networks Now Existing sensor network applications –Environmental/eco-system monitoring –Structural health –Agriculture Infrastructure-less environment Main design consideration –Energy efficiency

Vision Anticipated sensor network applications –Digital home, smart office –Healthcare –Workplace safety Mission-critical data Additional design considerations –Timely delivery –Reliable transmissions

Research Objective Data dissemination protocol –Timely delivery of data –Reliable transmission of data –Energy efficiency

Related Work Energy-efficient data dissemination –Cluster based –Probability based: random walk –Geographical based: location-aware Reliable data dissemination –Passive approaches error recovery –Active approaches Avoid congestion, selecting less lossy path This work aims at achieving timely delivery, reliability, and energy efficiency.

Magnetic Diffusion Consider the sink as a magnet Consider the data as metallic nails Two strategies of data propagation –Gradient-based (MDG) –Broadcast-based (MDB)

Gradient-based: Interest Broadcast Interest: data type, magnetic charge Sink

Gradient-based: Data Propagation Sending data according to gradients Sink Src

Broadcast-based: Interest Broadcast No gradients Sink

Broadcast-based: Data Propagation Data: magnetic charge, actual data Sink Src

Performance Evaluation Basic simulation setup Scenarios: static, mobile, on-off

Metrics Overhead –The amount of interest and data packet transmitted Reachability –The probability that the sink receives data successfully Latency –The data transmission time from the source to the sink

Two Sets of Comparisons I. Gradient-based vs. Broadcast-based –Which mode is better? II. MD vs. DD vs. Flooding –Is MD really better in terms of latency, reliability, and overhead? –Directed diffusion (DD) Two phase pull (TPP) and One phase pull (OPP)

I. Gradient-based vs. Broadcast-based

Overhead and Reliability MDB is more energy-efficient MDB is more reliable mobile case MDGMDB Interest # 9943 Data # Total # Reachability 80.67%86.27%

Latency MDB behaves better in latency –No handshake packets Thus, we adopt MDB for the rest of the comparison

II. MD vs. DD vs. Flooding

Total Overhead MD being multi-path, the overhead –No more than TPP –Much less than Flooding OPP TPP MD Flooding

Reachability In dynamic scenarios –Multi-paths give more reliable results –Multi-paths are not better in the static cases

Reachability with Random Wait Random wait mechanism –decreases the probability of collision

Latency in Static Scenario MD performs the best in latency –No handshake packets

Latency in Mobile Scenario MD is a better solution for applications with restricted latency requirement in dynamic network.

Latency in On-Off Scenario

Latency - Mobile with Random Wait This technique decreases the probability of collision, and in the meantime, increases transmission delay

System Selection Guideline static case dynamic case Static case –DD the best Dynamic case –MD better overall If 100% reliability is required –Flooding

Summary MD achieves in –Timely delivery –Reliability –Energy effectiveness –for dynamic sensor networks –An effective solution to mission-critical applications Simulation-based performance evaluation –Guidelines –for selecting the suitable mechanisms –for different application requirements

BL-Live: The TestBed

BL-Live –A mid-size sensor network testbed, 70+ sensor nodes –Transform BL Hall into a lively smart office building –Obtain practical experience and discover problems

BL-Live Hardware Two kinds of sensor nodes – Crossbow Micaz and Moteiv Telos. The placement –1 sink node in Lab 621 –2 sensor nodes with accelerometers in the elevators –72 relay nodes from the 4 th floor to the 6 th floor

BL-Live Services BL-Live provides two services: –Elevator Report –Smart Office

BL-Live Elevator Report Two slow paced elevators located on two opposite sides in BL Hall What if we can know the status of the elevator before we move to take?

BL-Live Elevator Report Sensor Networks Sink

Observations The reachability of MD is not good! (70+%) The reasons –Collisions Deployment is too dense MD broadcasts packets in multipath –Asymmetric links Link quality difference of A and B = |R ab -R ba |

Problem Caused By Asymmetric Links Sink A B C D There exists an asymmetric link!

Problem Caused By Asymmetric Links Sink A B C D Interest Broadcast

Problem caused by asymmetric links Sink A B C D Interest Broadcast

Problem Caused By Asymmetric Links Sink A B C D Data Propagation 6,data

Problem Caused By Asymmetric Links Sink A B C D This packet is lost. Data Propagation Node A won ’ t relay this pkt for node B. 7,data

Reliable Data Dissemination To improve the reliability of MD Counter two problems –Collision Random wait Priority –Two level forwarding Send Twice –Asymmetric Link MDlq MDfd

Random Wait Before sending the packet, it will wait for a random period of time. –Avoid collisions to increase the reachability

Priority Random wait increases the delay –Critical data need short latency Classify packets into two types –High priority and low priority Two-level Priority Forwarding –Send high priority packets first! To save queuing delay of high priority data

Send Twice –Send first copy immediately To shorten the latency –Send second copy in a random backoff To avoid the collision

Reliable Data Dissemination To improve the reliability of MD Counter two problems –Collision Random wait Priority –Two level forwarding Send Twice –Asymmetric Link MDlq MDfd

MDlq MD with revised interest broadcast method lq stands for link quality To set proper charge value for every node according to link quality

Revised Interest Broadcast Two phases –Link quality estimation CC2420 provides an indicator to estimate the link quality. –Interest Broadcast Specify the charge value and destination node id in the interest packets

Revised Interest Broadcast Sink A B C D A A A

Revised Interest Broadcast Sink A B C D A A A

Revised Interest Broadcast Sink A B C D A A A B B

Revised Interest Broadcast Sink A B C D A A,D,S A B B,C,S 7,A 6,B 6,C 5,S 5,D Interest broadcast phase is finished!

Revised Interest Broadcast Sink A B C D A A,D,S A B B,C,S The sink receives the data! 5,data 6,data 7,data

MDfd Everything is the same as MD, except … Send data with charge no larger than that of node Like flooding in a smaller area

MDfd Sink A B C D 6,data 7,data Node A will relay pkt for node B 7,data This data is received by sink.

MDlq V.S. MDfd MDlq –Advantage: Set proper charge value –Disadvantage: Overhead on revised interest broadcast MDfd –Advantage More paths –Disadvantage Overhead on new paths

Evaluation We want to see the impact of –To counter collision Random wait Priority –Two level forwarding Send twice –To counter asymmetric link MDlq MDfd All experimental data are collected in BL-Live

Experiment Setup Sink node1 Source node6 Relay node66 Period of interest broadcast 2 min. Period of critical data3 sec. Period of status data30 sec. Evaluation time80 min.

Evaluation Three metrics –Reachability –Latency –Overhead The amount of interest and data packets transmitted Highly related to energy consumption

Impact of Random Wait Reachability The reachability is increased by 5%.

Impact of Two Level Forwarding Latency Latency of high priority packet is slightly shorter.

Impact of Send Twice Reachability With send twice, the reachability is increased by 8%

Impact of MDlq and MDfd Reachability MDfd highly improves the reachability!

Overhead of MDlq and MDfd Overhead Overhead of MDfd very high.

MDlq+ Integrate different mechanisms –Increase the reachability –Energy efficient MDlq+ –MDlq with sendtwice

Impact of send twice MDlq+ and MDfd is close to Flooding! Reachability

Overhead of MDlq+, MDfd and Flooding Overhead MDlq+ is the most energy efficient.

Latency of MDlq+, MDfd and Flooding Latency of MDfd is as good as Flooding. MDlq+ is decent.

Summary of the Experimental Results Impact of –Random wait: increasing 5% –Two-level forwarding: Slightly shorten latency –Send twice Increasing 8% –Revised interest broadcast method Increasing 15% –MDfd and MDlq+ Close to Flooding(96%) More energy efficient

Summary Two Contributions –BL-Live Establish the testbed Manage the networking of sensor nodes –Reliable Data Dissemination Evaluate several mechanisms Improve the reachability to 95%

Questions?

Cross-Layer Analysis

– The standard in Wireless Network Contention-based protocol –RTS-CTS-DATA-ACK RTS CTS Sende r Receive r DATA ACK [Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specification, IEEE Std edition]

S-MAC - Periodic Listen and Sleep Contention-based protocol –RTS-CTS-DATA-ACK Listen interval –Send packets –Receive packets [W. Ye et al., “ An energy-efficient MAC protocol for wireless sensor networks ”, in INFOCOM 2002]

S-MAC – Schedule synchronization Schedules can differ –Neighboring nodes have same schedule Node 1 Node 2 sleep listen sleep listen sleep Schedule 2 Schedule 1 Border nodes: two schedules broadcast twice (Borrowed from S-MAC)

Scheduling in S-MAC Unknown neighbors –the same schedule Schedule 2 Schedule 1 Collision 1 Unicast Broadcast

B-MAC Contention-based protocol –No RTS/CTS, optional ACK Low Power Listening (LPL) –Preamble > Check-Interval [J. Polastre et al., Versatile low power media access for wireless sensor networks, Proceedings of the Second ACM Conference on Embedded Networked Sensor Systems (SenSys) 2004] Receive data Carrier sense Receiver Long PreambleData Tx Sender Check Interval (Borrowed from Z-MAC)

Low power listening (LPL) no RTS/CTS, optional ACK Schedule-based (TDMA ) Contention-based (CSMA) TDMA scheduling –Owners –non-owners [Injong Rhee, Ajit Warrier, Mahesh Aia and Jeongki Min, “ Z-MAC: a Hybrid MAC for Wireless Sensor Networks ”, ACM Sensys 2005] Z-MAC – On Top of B-MAC Hybrid (TDMA+CSMA)

Z-MAC – On Top of B-MAC Problem – hidden terminal collisions –Low contention level (LCL) –High contention level (HCL) Two-hop contention avoidance AB CD A Down

The Summarizations Non-energy efficient MAC – RTS-CTS-DATA-ACK Energy efficient MACs –S-MAC Periodic listen and sleep –B-MAC LPL, no RTS/CTS –Z-MAC: LPL TDMA + CSMA no RTS/CTS LCL/HCL

Experiments Simulation setup in NS2 simulator

Metrics Energy consumption –The amount of energy consumed in the network Reachability –The probability that the sink receives data successfully Latency –The data transmission time from the source to the sink

Energy Consumption MDB < MDG B-MAC best Z-MAC –TDMA scheduling MDB26800 MDG26804 Unit(J)

Energy Consumption – The Impact of Multiple sources Energy goes up –MDG-ZMAC –MDG-BMAC Overhead –MDB < MDG MDG + B-MAC MDG + Z-MAC MDB+ Z-MAC MDB + B-MAC

Energy Consumption - Summarization Energy consumption –MDB < MDG –B-MAC < S-MAC < Z-MAC < –Best - MDB + BMAC LPL is sensitive to the traffic load Routing and MAC –Critical to the energy consumption

Reachability In –MDB < MDG In S-MAC –MDB > MDG Sink Source MDG MDB Down

Reachability – The Impact of Multiple Sources High traffic load –MDG –MDG + Z-MAC MDG + Z-MAC MDG

Reachability - Summarization The relative performance of routing protocols changes –When run over different MACs In dense network –S-MAC is bad Reachability –Retransmission –Two-hop collision avoidance –MDG and MDG + Z-MAC

Latency MDB-802 best MDB-BMAC –Delay < 1 sec –80% < 500ms MDB MDB + B-MAC

Latency - Summarization Generally speaking, MDB is better The relative performance is not obvious Latency –MDB is the best –MDB + B-MAC is surprisingly good –Delay can be short In an energy-efficient MAC

System Selection Guidline The selection of protocol combination depends on –Application –Deployment environment Elevator application in BL-Live [Seng-Yong Lau et al., “Sensor Networks for Everyday Use: The BL-Live Experience“, IEEE International Conference on Sensor Networks, Ubiquitous, and Trustworthy Computing (SUTC 2006)]

Summary The interactions between routing and MAC –Relative performance might change –Both are critical to energy consumption –No one wins in every case High reliability in an energy-efficient MAC –Retransmission –Two-hop collision avoidance

Contribution We achieves in –Cross-layer performance evaluation Relative performance might change The interaction between routing and MAC In wireless sensor network –System selection guidelines

Future Work Extensive set of experiments Various routing protocols Real test-bed

Questions?