Many Eyes: Projections for Sensor Networks

Keynote talk at MDM 2004. A tasting of technical challenges and societal questions for sensor networks. Focus on the multi-layer optimization problem of computing different classes of functions over sensornet topologies. Also summarizes recent discussions on legal implications of sensornet privacy, joint with Boalt School of Law at UC Berkeley.

Transcript

1. projections for sensor networks joseph m hellerstein uc berkeley intel research berkeley many eyes
2. road map the romantic view research experiences & opportunities the spanish inquisition? societal implications: privacy and the law
3. “But O the truth, the truth! the many eyes That look on it! the diverse things they see!” George Meredith, “A Ballad of Fair Ladies in Revolt”
4. “But O the truth, the truth! the many eyes That look on it! the diverse things they see!” George Meredith, “A Ballad of Fair Ladies in Revolt”
5. road map for romance a tour of current technology deployments present and near future a tasting of research challenges
6. what is a sensor net large collection of wireless compute/sense nodes self-organized into a network sensor node: microprocessor, radio, memory, sensors temp, humidity, pressure, vibration, magnetics, sound, etc. many/tiny for dense spatial sampling 1000’s of nodes
7. what is a sensor net large collection of wireless compute/sense nodes self-organized into a network sensor node: microprocessor, radio, memory, sensors temp, humidity, pressure, vibration, magnetics, sound, etc. many/tiny for dense spatial sampling 1000’s of nodes
8. big deal? just a little computer with a radio! sensors are already a big industry. but: wireless qualitatively changes deployment many/tiny qualitatively changes use cases significant research challenges due to many/tiny, wireless (battery), ad-hoc
9. big deal? just a little computer with a radio! sensors are already a big industry. but: wireless qualitatively changes deployment many/tiny qualitatively changes use cases significant research challenges due to many/tiny, wireless (battery), ad-hoc
10. big deal? just a little computer with a radio! sensors are already a big industry. but: wireless qualitatively changes deployment many/tiny qualitatively changes use cases significant research challenges due to many/tiny, wireless (battery), ad-hoc
11. big deal! research depth and breadth emerging ACM SenSys, IPSN DARPA SensIT, HSARPA, NSF, etc. startups Crossbow, Dust, Ember, MillenialNet, ... media buzz Wired, Technology Review, etc.
12. to deploy lots and lots of these, must be: cheap capable of ad-hoc, multi-hop networking easy to deploy and task en masse zero-admin (pref. disposable) long-lived (months-years) requirements
13. the berkeley/intel suite overview of field from one perspective motes TinyOS & NesC TinyDB TASK
14. Mica-2 Dot ~ 1980s PC-AT 8MHz ATMega 128L 4K code segment 128K RAM / 512K flash 900 MHz radio stackable sensors berkeley motes image: www.xbow.com
15. power is key. (comm is main draw) further miniaturization on its way future motes image: www.dust-inc.com
16. power is key. (comm is main draw) further miniaturization on its way future motes image: www.dust-inc.com
17. power is key. (comm is main draw) further miniaturization on its way future motes image: www.dust-inc.com
18. iMote an intel research prototype bluetooth network image: www.intel.com/research/exploratory/motes.htm
19. stargate a palmtop-class gateway/appliance intel XScale + lots of connectivity image: www.xbow.com
20. standards efforts IEEE 802.15.4 low-power wireless radio open standard for physical & MAC layer berkeley TELOS and Crossbow MICA-Z Zigbee alliance closed standard defining NW, sec, app layers
21. NesC/TinyOS
22. NesC/TinyOS NesC: a C dialect for embedded programming components, “wired together” quick commands and asynch events
23. NesC/TinyOS TinyOS: a set of NesC components hardware components ad-hoc network formation & maintenance time synchronization NesC: a C dialect for embedded programming components, “wired together” quick commands and asynch events
24. NesC/TinyOS TinyOS: a set of NesC components hardware components ad-hoc network formation & maintenance time synchronization think of the pair as a programming environment NesC: a C dialect for embedded programming components, “wired together” quick commands and asynch events
25. TinyOS status TinyOS 1.1.1 released november, 2003 by end-of-year over 8,000 downloads 1.1.3 released this month runs on berkeley motes & imotes
26. to deploy lots and lots of these, must be: cheap capable of ad-hoc, multi-hop networking easy to deploy and task en masse zero-admin (pref. disposable) long-lived (months-years) requirements
27. to deploy lots and lots of these, must be: cheap capable of ad-hoc, multi-hop networking easy to deploy and task en masse zero-admin (pref. disposable) long-lived (months-years) requirements
28. programming sensornets
29. programming sensornets distributed and embedded programming
30. programming sensornets distributed and embedded programming data oriented
31. programming sensornets distributed and embedded programming data oriented a new metaphor: real world as a database seshadri/bonnet at cornell
32. querying the world part of a bigger nets/dbs agenda theme: declarative programming for large, unpredictable networks of machines see also p2p work like chord, bamboo, pier, etc. codd’s data independence, recast
33. TinyDB: the model data model: sensornet as a table row per node attribute per sensor type logical attributes: sensorID, time, etc. query model: continuous SQL based on epochs Epoch SensorI D Time Temp Light Magneto 0 2 14:03:56.3 68.4 81.8 21.1 0 1 14:03:56.6 67.8 79.1 74.2 0 3 14:03:56.8 84.2 98.3 31.6 1 1 14:04:26.4 67.8 79.0 21.4
34. TinyDB: the model data model: sensornet as a table row per node attribute per sensor type logical attributes: sensorID, time, etc. query model: continuous SQL based on epochs Epoch SensorI D Time Temp Light Magneto 0 2 14:03:56.3 68.4 81.8 21.1 0 1 14:03:56.6 67.8 79.1 74.2 0 3 14:03:56.8 84.2 98.3 31.6 1 1 14:04:26.4 67.8 79.0 21.4 (per epoch)
35. TinyDB: the model data model: sensornet as a table row per node attribute per sensor type logical attributes: sensorID, time, etc. query model: continuous SQL based on epochs Epoch SensorI D Time Temp Light Magneto 0 2 14:03:56.3 68.4 81.8 21.1 0 1 14:03:56.6 67.8 79.1 74.2 0 3 14:03:56.8 84.2 98.3 31.6 1 1 14:04:26.4 67.8 79.0 21.4 (per epoch) implicit acquisitional model
36. TinyDB extensibility extensible attributes add a new sensor (e.g. soil moisture) add a new logical attribute (e.g. room number) extensible functions & commands user-defined aggregates (e.g. wavelet) user-defined functions/commands (e.g. sound buzzer)
37. a TinyDB query SELECT roomID, AVG(temp) FROM sensors GROUP BY roomID EPOCH DURATION 300
38. TinyDB execution pattern “flood” query to all nodes a tree is formed based on arrival pattern periodically communicate up the tree data reduction opportunities tree reconfigures itself A B C D E F
39. A B C D E F Q Q TinyDB execution pattern “flood” query to all nodes a tree is formed based on arrival pattern periodically communicate up the tree data reduction opportunities tree reconfigures itself
40. A B C D E F Q Q Q Q R:{...} R:{...} TinyDB execution pattern “flood” query to all nodes a tree is formed based on arrival pattern periodically communicate up the tree data reduction opportunities tree reconfigures itself
41. A B C D E F R:{...} R:{...} R:{...} R:{...} Q Q Q TinyDB execution pattern “flood” query to all nodes a tree is formed based on arrival pattern periodically communicate up the tree data reduction opportunities tree reconfigures itself
42. A B C D E F R:{...} R:{...} R:{...} R:{...} R:{...} TinyDB execution pattern “flood” query to all nodes a tree is formed based on arrival pattern periodically communicate up the tree data reduction opportunities tree reconfigures itself
43. expressivity this is a restricted communication pattern yet quite useful epochs predictable, battery-efficient extensible agg yields many useful apps object tracking, isobar mapping, wavelet compression, etc. repeating the pattern yields a family of statistical & coding algorithms via junction trees bayesian belief propagation, fft’s, regression, turbo decoding, etc.
44. sensornet in a box stargate, motes software: Tiny Application Sensor Kit mote-side: TinyDB with fixed schema stargate: TinyDB Java clients, PostgreSQL database, Apache webserver configure via any web browser TASK
45. A B C D E F sensornet in a box stargate, motes software: Tiny Application Sensor Kit mote-side: TinyDB with fixed schema stargate: TinyDB Java clients, PostgreSQL database, Apache webserver configure via any web browser TASK
46. road map for romance technology status report deployments present and near future a flavor of research challenges
47. deployments
48. deployments storm petrel habitat monitoring, great duck island
49. deployments storm petrel habitat monitoring, great duck island redwood forest microclimate monitoring
50. deployments storm petrel habitat monitoring, great duck island redwood forest microclimate monitoring golden gate bridge vibration monitoring
51. deployments storm petrel habitat monitoring, great duck island redwood forest microclimate monitoring golden gate bridge vibration monitoring still to come intel fab vibration hp smart data center sap asset tracking riverbed motion
52. data from the redwoods
53. road map for romance technology status report deployments present and near future a flavor of research challenges
54. queries on networks a multi-layer optimization problem with: error/loss tolerance approximation online adaptivity
55. optimization layers physical connectivity
56. optimization layers physical connectivity routing tree
57. optimization layers physical connectivity support graph f() routing tree
58. optimization layers physical connectivity routing tree support graph f’() support graph f()
59. optimization layers physical connectivity routing tree support graph f’() support graph f() loss mitigation
60. optimization layers physical connectivity routing tree support graph f’() support graph f() loss mitigation adapt to changing physical layer
61. an example: wavelets re-codes a signal into a sum of scaled/shifted basis functions haar wavelet based on simple square waves biggest coefficients 㱺 approximate reconstruction lossy compression
62. wavelet histograms 2-d or 3-d (spatio-temporal) compression for reconstruction a big picture of the data
63. haar support graph — a b — c d — — — + + + + — e f — g h + + — i j — k l — — — + + + + — m n — o p + + + + + and one possible communication graph
64. haar support graph — a b — c d — — — + + + + — e f — g h + + — i j — k l — — — + + + + — m n — o p + + + + + and one possible communication graph
65. haar support graph — a b — c d — — — + + + + — e f — g h + + — i j — k l — — — + + + + — m n — o p + + + + + and one possible communication graph
66. haar support graph — a b — c d — — — + + + + — e f — g h + + — i j — k l — — — + + + + — m n — o p + + + + + and one possible communication graph
67. haar support graph — a b — c d — — — + + + + — e f — g h + + — i j — k l — — — + + + + — m n — o p + + + + + and one possible communication graph
68. haar support graph — a b — c d — — — + + + + — e f — g h + + — i j — k l — — — + + + + — m n — o p + + + + + and one possible communication graph
69. resulting comm graph a b c d e f g h i j k l m n o p
70. resulting comm graph a b c d e f g h i j k l m n o p a binomial tree!
71. continuing the fun probability of a good binomial comm graph at physical layer? tradeoff requiring a binomial tree against coping tradeoff against approximate versions of haar loss tolerance online adaptivity
72. road map the romantic view research experiences & opportunities the spanish inquisition? societal implications: privacy and the law
73. “Fear has many eyes, and sees things underground. All the moreso in heaven above.” Miguel de Cervantes, Don Quixote
74. “Fear has many eyes, and sees things underground. All the moreso in heaven above.” Miguel de Cervantes, Don Quixote
75. two caveats i am not an expert on the legal aspects of privacy or any other legal issues summary of conversations with pam samuelson and deirdre mulligan at boalt (ucb law ) what i will describe is very US-centric
76. privacy concerns mount RFID already causes significant concern benneton wal-mart/gillette recent verisign “ons” announcement
77. legal “devices” for privacy decisional law e.g., discussions w/your doctor or lawyer search & seizure (4th amendment) also trespass law information privacy confidentiality of communications
78. some privacy law b.g. brandeis/warren article in h.l.r. 1890 “right to be left alone” based on paparazzi at warren family wedding b&w used other aspects of the law to justify law is based on normative behavior technology can cause norms to change
79. some privacy law b.g. reproductive privacy griswold v. conn. (1965): constitutional “penumbra” right to privacy roe v. wade (1975) said right to privacy “fundamental” katz v. u.s. (1967) gov’t wiretapping on telephone booth extended 4th amendment protection privacy based on person, not place test for balancing privacy against interest in protecting society
80. ftc fair info practices in the electronic marketplace: notice (what, how, why, for whom) choice (on 2ndary use) access (esp. for review and correction) security (website owner must secure data) (missing: minimization) how does this map to sensornets or any ubiquitous computing technology?
81. privacy is contextual place: patient/doctor conversation in a restaurant info: tell your doctor you committed murder person: tell your boss your medical woes currently: special case law for: video rentals, driver’s license data, financial, health, etc. implications for ubicomp and sensornets?
82. etudes in sensornet privacy imagine trying to enforce or audit sensornet use in the spirit of fair information practices what cost fair play? imagine trying to subvert sensors detect, evade, jam, confuse, destroy, etc. info chaff vary spatial and temporal granularity
83. one message: engage privacy concerns are real legal and technical “devices” can be employed norms are determined (and broken!) by technologists and users now is the time to talk technologists, ethnographers, legal experts, gov’t, industry
84. “Fear has many eyes, and sees things underground. All the moreso in heaven above.” “But O the truth, the truth! the many eyes That look on it! the diverse things they see!”
85. “Fear has many eyes, and sees things underground. All the moreso in heaven above.” “But O the truth, the truth! the many eyes That look on it! the diverse things they see!”
86. “Fear has many eyes, and sees things underground. All the moreso in heaven above.” “But O the truth, the truth! the many eyes That look on it! the diverse things they see!”
87. so much is open here enormous breadth of ideas at play technical work needed at all levels architectural algorithmic interfaces and languages applications huge potential for impact both positive and negative
88. come explore open platforms available hobbyist fun: like early days of computing room for many eyes and hands and minds
89. further info http://telegraph.cs.berkeley.edu/tinydb http://webs.cs.berkeley.edu http://www.intel-research.net/berkeley
90. backup slides