CMPT 401 Summer 2007 Dr. Alexandra Fedorova Lecture XVI: Mobile and Ubiquitous Computing
Feb 10, 2016
CMPT 401 Summer 2007
Dr. Alexandra Fedorova
Lecture XVI: Mobile and Ubiquitous Computing
2CMPT 401 Summer 2007 © A. Fedorova
Mobile and Ubiquitous Computing
• Mobile computing – computers that users can carry – Laptops, handhelds, cell phones– Wearable computers
• Heart monitors used by athletes (Tour de France: team manager monitors heart rates, give recommendations on tactics)
• Health monitors used by elderly
• Ubiquitous computing– Computers are everywhere– Each person uses more than one computer– PC, laptop, cell phone, watch, car computer (100+
microprocessors in some cars)
3CMPT 401 Summer 2007 © A. Fedorova
Enables New Cool Applications
• Object tracking– Track location of a child, parent, dog, car (lojack)– Parents watch their babies in the daycare
• Health monitoring– Monitor child breathing (prevent SIDS – sudden infant death syndrome)– Heart stimulation: embed hearth sensors in the elderly. If pulse goes too low,
stimulate the pulse• Replace physicians visits (Neuromancer project at Sun Microsystems, Jim
Waldo)– People wear health monitors– They collect health data normally measured by doctors/nurses– Eliminates the need for doctor visits – sensors can alert of dangerous health
conditions– Massive data available – a chance to carry out longitudinal studies in medicine
4CMPT 401 Summer 2007 © A. Fedorova
Some Challenges
• Limited power– Wearable devices and sensors have low battery power– To be interesting, sensors must transmit data– Data transmission uses power– How to minimize power consumption and maximize transmission of
useful data?• Limited network bandwidth
– Applications must communicate to sensors exactly what data they need, so sensors don’t transmit useless data
• Limited connectivity– Mobile devices often operate in disconnected mode– How to associate to a new network seamlessly?– How to form a network without an infrastructure (ad-hoc networking)?
5CMPT 401 Summer 2007 © A. Fedorova
More Challenges
• Sensor deployment– Sensors have limited lifetime, at some point they become useless – In ecologically sensitive environments – this means a bunch of silicon
scattered around– Example: deploy sensors for forest fire detection. Scatter sensors around
the forest (from a helicopter)– After a while you have a whole lot of improperly disposed batteries
• Handling data– Once all these super-apps get implemented, we’ll have massive amounts of data
collected by all imaginable sensors– Much of this data will be kept around for historical analysis– Where do we store this data? (P2P? – addressed by Neuromancer)– How do we make sure it’s safe (replication?)– How do we make sure it’s secure?
6CMPT 401 Summer 2007 © A. Fedorova
Case Studies of Sensor Networks
• Design and Deployment of Industrial Sensor Networks: Experiences from a Semiconductor Plant and the North Sea, Krishnamurthy et al.
• IrisNet: An Architecture for a Worldwide Sensor Web, Gibbons et al.
7CMPT 401 Summer 2007 © A. Fedorova
Industrial Sensor Networks• Sensor networks used for predictive equipment maintenance
– Monitor industrial equipment– Detect oncoming failures– Alert humans of potential failures
• We will talk about– Motivation– System architecture– System issues specific to wireless sensor networks
• Two case studies– Semiconductor fabrication plan– Oil tanker in the North Sea
8CMPT 401 Summer 2007 © A. Fedorova
Predictive Equipment Maintenance (PdM)
• Monitor and assess the health status of a piece of equipment (e.g., a motor, chiller, or cooler)
• PdM allows to detect most failures in advance• But analysis has to be performed with sufficient frequency• Equipment has sensors attached to it• Sensors monitor conditions of the equipment• Report results to the operator’s computer• Operator analyses data, detects any unusual patterns,
decides if failure is imminent• Takes action to replace the equipment
9CMPT 401 Summer 2007 © A. Fedorova
Types of Sensor Data• Vibration (used in this study) – analyze frequency and amplitude of vibrations
over time– Identify unexpected changes – suggest repair or replacement– Source of vibrations must be identified and assigned to a specific component
• Oil analysis – analysis of wear particles, viscosity, acidity and raw elements– Capture a small sample, compare to baseline samples – detect potential
problems• Infrared Thermography – sense heat at frequencies below visible light
– Detect abnormal heat sources, cold areas, liquid levels in vessels, escaping gases
• Ultrasonic detection – detect wall thickness, corrosion, erosion, flow dynaics, wear patterns
– Compare data to standard change rates, project equipment lifeime
10CMPT 401 Summer 2007 © A. Fedorova
Importance of PdM
• Reduce catastrophic equipment failures• Save human lives• Reduce associated repair and replacement cost• Save money – switch from calendar-based maintenance to
indicator driven maintenance– Calendar-based maintenance: may do maintenance when you don’t
need to – May fail to do the maintenance when you really have to
• Quantify the value of a new system within the warranty period
• Meet factory uptime and reliability requirements
11CMPT 401 Summer 2007 © A. Fedorova
Existing PdM Technologies: Manual Data Collection
A human operator visits the equipment under surveillance
Sensors are installed in the equipment or brought by the operator
Data is transported to the lab for analysis
Data is collected into a hand-held device
12CMPT 401 Summer 2007 © A. Fedorova
Existing PdM Technologies: Online Surveillance
Sensors are connected to equipment, hardwired to data acquisition unitData acquisition unit processes the data and delivers it across a wired network to a central repository
SensorData acquisition
unit
Central repository
13CMPT 401 Summer 2007 © A. Fedorova
Disadvantages of Existing Technologies
• Manual data collection:– Potential for user error– High cost to train and keep experts– Cost of manpower for frequent data collection– Most users of manual data collection are not happy with the level
of prediction and correlation• Online systems:
– Cost of hardware and network infrastructure– Only appropriate for equipment with cost impact of over $250K in
case of failure– Online systems are used in only 10% of the market (due to cost)
14CMPT 401 Summer 2007 © A. Fedorova
Wireless Sensor Networks for PdM
• Provide frequency of monitoring comparable to online systems
• Lower cost of deployment – network is wireless– Just drop the sensors and you are ready to go
• Data acquisition unit needs not be specialized hardware– Just any computer that can listen for radio signals from sensors
15CMPT 401 Summer 2007 © A. Fedorova
Challenges in Deployment of Wireless Sensor Networks
• Determine requirements for industrial environments:– How often does data need to be sampled? – In what form to transmit and organize the data?– How long will the sensor battery survive?
• Effect of environment on deployment– What is the signal quality in the current environment? Lots of thick
walls is bad for the signal– How often will the network be disconnected – i.e., in the ship the
compartment containing sensors is periodically shut off• How to ensure the required quality
– Sensors will fail, how do you ensure that sufficient data collection rates are achieved?
16CMPT 401 Summer 2007 © A. Fedorova
Setup for Vibration Analysis
• Accelerometer – a device used to measure vibrations or accelerations due to gravity change or inclination
• Measures its own acceleration, so it must be hard-mounted to themonitored equipment
• In the experiment, an off-the-shelf accelerometer was used; it interfaces with the rest of the sensor board (radio, etc.)
• Sensor network interfaces with an off-the-shelf software application – provides long term data storage, trend analysis, fault alarms
17CMPT 401 Summer 2007 © A. Fedorova
Site Planning
• How/where to install the sensors given the particularities of a given site?
• Sensors must be safe for the equipment they monitor• Radio Frequency (RF) coverage – are there walls and
equipment preventing good RF coverage? Must relay nodes or gateways be installed?
• RF interference – is there RF noise that will prevent good transmission? RF interference may come from other radios used on the site.
• To assess these factors, a site survey is needed
18CMPT 401 Summer 2007 © A. Fedorova
Site Survey
• Place test sensors near sensing points (where actual sensors will be mounted in the future)
• Place test gateways (the machines that will receive data from sensors and transmit it further) at locations where actual gateways would be placed– Near power outlets and Ethernet jacks
• Using test setup, evaluate wireless connectivity, RF coverage and interference
19CMPT 401 Summer 2007 © A. Fedorova
Site Survey Results• Sensor nodes with more powerful radios worked better in
conditions with RF interference• Less powerful radios were not able to transmit through a
door on the oil tanker• It had to be ensured that sensor node frequencies did not
overlap with critical radio frequencies used on the oil tanker• Witnessed better RF performance on the oil tanker than was
initially expected:– Attributed to use of steel materials on the ship– Steel materials reflect, rather than attenuate RF energy (unlike office
and home environments)
20CMPT 401 Summer 2007 © A. Fedorova
Application Specific Requirements
• Data must be accurate, acquired and transmitted in a timely manner– Challenge: sensors and data acquisition units will fail due to
operation in a harsh environment– Solution: system must be designed with expectation for failure
and with ability to quickly recover from failures• Long-lived battery powered operation
– Sensor networks should not use plant power – Should be battery operated: must operate for a long time on one
set of batteries, to avoid the need for frequent redeployment
21CMPT 401 Summer 2007 © A. Fedorova
Hardware Architecture
• Two types of sensor nodes :– Mica2 Mote– Intel Mote
• Mote:– Composed of a small, low
powered computer– Radio transmitter– Connected to several
sensors• The node’s sensor board is
connected to vibration sensors
Sensor node (Mica2 mote)
22CMPT 401 Summer 2007 © A. Fedorova
Hardware Architecture Comparison
• Mica2– Less powerful radio– No on-board storage for sensor data, so you need to attach
additional storage to it• Intel
– Very powerful radio: 10x throughput of the Mica2 mote– Uses more power
23CMPT 401 Summer 2007 © A. Fedorova
Network Architecture
• Hierarchical architecture– Sensor clusters (sensor
mesh)– Cluster head (connected to
the gateway)– Stargate Gateway
• mote radio• 802.11 radio
– 802.11 backbone– Root Stargate– Bridge Stargate– Enterprise server
24CMPT 401 Summer 2007 © A. Fedorova
Data Collection and Transfer
• Cluster head schedules data capture/transfer for every sensor connected to each node
• When a node has captured data it initiates a connection to the Stargate gateway
• Data is transferred using a reliable transport protocol• Sensor data is time-stamped and put in a file• There is a separate file for each collection of a sensor channel• Each Stargate gateway periodically copies file to the root gateway• Root gateway transfers data to Bridge gateway via serial cable – this
is done to isolate wireless network from the corporate network• Bridge gateway transfers data to the enterprise server
25CMPT 401 Summer 2007 © A. Fedorova
Hierarchical Network Structure
• Tier 1 – lowest level– Networks of sensor nodes– They form clusters: may be pre-assigned to a cluster or choose the cluster
dynamically– Lowest compute capability, limitations on bandwidth and battery capaciry
• Tier 2 – middle level– Sensor network backbone– Individual cluster gateways– Higher compute and power capacity – offload computational burden from
Tier 1• Tier 3 – highest level
– Interface to the enterprise– Abstracts application needs from the sensor network
26CMPT 401 Summer 2007 © A. Fedorova
Sleep/Wakeup Schedule
• Sensor nodes form a cluster around a gateway• Nodes in a cluster follow a sleep/wakeup protocol• When nodes wake up they acquire data from sensors and
transmit it to the gateway• Then they go to sleep until the next data collection is
scheduled• Sleep/wake-up operation saves battery power• Sleep/wake-up schedule is coordinated by a cluster head
– a device connected to the gateway via a serial port
27CMPT 401 Summer 2007 © A. Fedorova
Power Management Protocol
• Cluster head schedules sleep periods based on application-level sampling requirement
• Upon initial discovery of nodes in the cluster, cluster head sends the first request for data collection
• At the end of each data collection it sends a message indicating start time and duration of next sleep phase
• Sensor nodes go to sleep and then wake up all together• When nodes are asleep they are not completely turned off, but they
operate in a low power mode• Nodes’ clocks are not perfectly synchronized, so the cluster head
waits for some “skew” period until beginning next data collection• Sleep periods in the oil tanker installation were set to 7 and 18 hours
28CMPT 401 Summer 2007 © A. Fedorova
Fault Tolerance
• Sensor networks must operate in harsh environments for long periods of time
• Failures are common and should be expected
29CMPT 401 Summer 2007 © A. Fedorova
Fault Tolerant Design
• Four design features to increase fault tolerance:– Watchdog timers – a node resets itself upon encountering
unexpected behavior– Cluster heads store network state – nodes can return to operation
quickly after being reset– Intentional re-initialization of sensor nodes after each collection
period– Non-volatile storage of critical state at cluster head – cluster head
could be (and was) reset after each wake-up period
30CMPT 401 Summer 2007 © A. Fedorova
Watchdog Timers
• Each node monitors events:– How much time has passed since last packet reception (in the
wake state)– Events signifying radio lockups– Protocol events – e.g., receipt of new data send request before
the previous one was finished• The node resets itself if any of these unexpected events
was detected
31CMPT 401 Summer 2007 © A. Fedorova
Comparing Power Consumption
• Active power – power when the network is awake– Similar usage of active power per unit of time– But Intel motes spent less time being awake, because they had faster radios– So Intel-based network used less power overall
• Power during the sleep phase– Intel network implemented a connected sleep mode– You can still access the network while the nodes are asleep, albeit at a
higher latency– So it used more power in the sleep mode– If Intel-based network were completely disconnected, it would use only
slightly more power as Mica2-based network– Using an external real-time clock can enable completely turning off the
network during the sleep mode – even more power would be saved
32CMPT 401 Summer 2007 © A. Fedorova
Battery Life
• On the oil tanker, two lengths of sleep mode were used:– 18 hour sleep period– 5 hour sleep period
• Resultant battery lives are:– 18-hour period: 82 days– 5-hour period: 21 days
33CMPT 401 Summer 2007 © A. Fedorova
Case Studies of Sensor Networks
• Design and Deployment of Industrial Sensor Networks: Experiences from a Semiconductor Plant and the North Sea, Krishnamurthy et al.
• IrisNet: An Architecture for a Worldwide Sensor Web, Gibbons et al.
34CMPT 401 Summer 2007 © A. Fedorova
IrisNet
• A slightly different environment than conventional sensor networks
• Many devices: PCs, hand-helds, cameras• Good connectivity, no power limitations• Provide useful data• Question:
– How do we access and integrate this data to enable interesting applications?
• Solution: – Architecture for a Worldwide Sensor Web
35CMPT 401 Summer 2007 © A. Fedorova
IrisNet Vision
• A user will query, as a single unit, vast quantities of data from thousands of widely distributed sensors
• Many possible uses:– Epidemic Early Warning System - monitor water quality, oil spills– Homeland Security– Computer Network Monitoring – gather (sense) data on
bandwidth/CPU usage; answer queries such as “What’s the least loaded node at SFU?”
– Traffic / Parking Assistance – help me find hockey game parking in Vancouver
36CMPT 401 Summer 2007 © A. Fedorova
IrisNet Goals• Planet-wide local data collection and storage
– Massive amounts of data– Retain data near its source, transmit to the Internet only as needed
• Ease of service authorship– Vision: when sensors are deployed, we don’t know all potential users– Different service providers might want to collect different data and different rates
and apply different filters depending on the service• Real-time adaptation of collection and processing
– Reconfigure data collection and data filtering processes, change sampling rates• Data as a single queriable unit
– Global sensing device network is a single unit that supports a high-level query language
– Users make complex queries: “Tell me the location of my grandmother at the time when the oil spill in the Baltic sea was first detected”.
• Data integrity and privacy– No one should be able to query my health data except my doctor
37CMPT 401 Summer 2007 © A. Fedorova
IrisNet Architecture
User
. . .Sensor
SA
senseletsenselet
Sensor
SA
senseletsenselet
Sensor Sensor
SA
senseletsenselet
Web Serverfor the url
. . .Query
OAXML database
. . .OA
XML databaseOA
XML database
Two components:SAs: sensor feed processingOAs: distributed database
From slides of P. Gibbons
38CMPT 401 Summer 2007 © A. Fedorova
IrisNet Architecture
• Sensing Agents (SA)– Generic data acquisition interface: ask sensor to collect data X at
frequency Y, filter data according to parameter Z– A service configures sensing agent according to its needs– Configuration is done via execution of service-specific code
senslet– A single SA can execute one or more senslets
• Organizing Agents (OA)– Service specific sensing data is stored in a database– This database is queried by users
39CMPT 401 Summer 2007 © A. Fedorova
Organization of SA
40CMPT 401 Summer 2007 © A. Fedorova
OA Architecture
• XML-based database• Hard to design rich schema
for all possible service• XML allows the use of self-
describing tags• Database is partitioned and
distributed• Replicate parts of the
database• Primary replicas: strong
consistency• Secondary replicas: weak
consistency
41CMPT 401 Summer 2007 © A. Fedorova
Querying the IrisNet
• Each node has a human readable name• Each such name is registered in the DNS with associated IP address• Query is routed to the IP address
neighbourhood-Shadyside.
city-Pittsburgh.state-PA.
usRegion-NE.intel-iris.net
42CMPT 401 Summer 2007 © A. Fedorova
Example Services
• Parking space finder– Uses cameras throughout a metropolitan area to track parking space availability– Users fill out a Web form to specify destination and any constraints on a desired
parking space– Parking space finder identifies the parking space satisfying constraints
• Network and host monitor (IrisLog)– Collects data from computer and network monitoring tools– Those tools act like sensors– They report data, such as CPU and memory load, network bandwidth– Answer queries such as “find the least loaded node on the network”
• Coastal imaging service– Uses camera installed at Oregon coastline– Uses live feed from cameras to identify signatures of phenomena such as riptides
and sandbar formations
43CMPT 401 Summer 2007 © A. Fedorova
Summary
• Variety and quantity of small computers is exploding• These computers are mobile, wearable, provide a variety of cool
functions/sensing abilities, and are affordable!• One can imagine a multitude of useful “killer apps” using those
devices• Many challenges need to be overcome to make these applications
really work:– Limited power and network bandwidth– Formation of ad-hoc networks– Querying the available data– Handling and storing massive amounts of data