Opening the Sensornet Black Box. Megan Wachs, Jung Woo Lee, Jung Il Choi, & Philip Levis Computer Systems Laboratory, Stanford University

Transcription

1 Opening the Sensornet Black Box Megan Wachs, Jung Woo Lee, Jung Il Choi, & Philip Levis Computer Systems Laboratory, Stanford University

2 The Sensornet Black Box It is difficult to observe what occurs deep within a sensor network This is the direct result of energy constraints on a mote. This lack of visibility is the principal difficulty in development.

3 The MNet Architecture Minimize the energy cost of diagnosing the cause of a failure or behavior

4 Outline Survey of Failures The MNet Architecture A Design Example Further Implications

5 What kinds of failures are observed in real deployments? System Interactions: software conflicts Murphy Loves Potatoes Vigilnet A Line in the Sand Unwired Wine

6 What kinds of failures are observed in real deployments? System Interactions: software conflicts Network Problems: Saturation & Congestion A Line in the Sand TASK Vigilnet Sensorscope The Heathland Experiment Flush

7 What kinds of failures are observed in real deployments? System Interactions: software conflicts Network Problems: Saturation & Congestion Protocol Issues: Conflicts & Failures Murphy Loves Potatoes

8 What kinds of failures are observed in real deployments? System Interactions: software conflicts Network Problems: Saturation & Congestion Protocol Issues: Conflicts & Failures Unknown Unwired Wine TASK Murphy Loves Potatoes The Heathland Experiment Monitoring Volcanic Eruptions Industrial Sensor Networks A Line in the Sand

9 Effects of Failures on Deployment Performance Great Duck Island: 58% Peter Scott R. Szewczyk, J. Polastre, A. Mainwaring, and D. Culler. An analysis of a large scale habitat monitoring application. In Proceedings of the Second ACM Conference On Embedded Networked Sensor Systems (SenSys), 2004.

10 Effects of Failures on Deployment Performance Great Duck Island: 58% Redwoods : 40% G. Tolle, J. Polastre, R. Szewczyk, D. Culler, N. Turner, K. Tu, S. Burgess, T. Dawson, P. Buonadonna, D. Gay,, and W. Hong. A macroscope in the redwoods. In Proceedings of the Third ACM Conference on Embedded Networked Sensor Systems (SenSys), 2005.

11 Effects of Failures on Deployment Performance Great Duck Island: 58% Redwoods : 40% Potato Field: 2% K. Langendoen, A. Baggio, and O. Visser. Murphy loves potatoes: Experiences from a pilot sensor network deployment in precision agriculture. In the Fourteenth Int. Workshop on Parallel and Distributed Real-Time Systems (WPDRTS), 2006.

12 Effects of Failures on Deployment Performance Great Duck Island: 58% Redwoods : 40% Potato Field: 2% Volcan Reventador: 68% G. Werner-Allen, K. Lorincz, J. Johnson, J. Leess, and M. Welsh. Monitoring volcanic eruptions with a wireless sensor network. In Proceedings of the Second European Workshop on Wireless Sensor Networks (EWSN), 2005.

13 Management and Debugging Sympathy Lightweight RPC Network Snooping Tools

14 Management and Debugging Sympathy Lightweight RPC Network Snooping Tools Sympathy

15 Management and Debugging Sympathy Lightweight RPC Network Snooping Tools Sympathy

16 Outline Survey of Failures The MNet Architecture A Design Example Further Implications

17 Visibility Metric Minimize the energy cost of diagnosing the cause of a failure or behavior

18 The Need For Isolation Isolation simplifies reasoning.

19 The Need For Isolation Isolation simplifies reasoning. Isolation alone is not enough, the network must also provide fairness.

20 The Need For Isolation Network Isolation between protocols: Please be quiet!

21 Fair Waiting Protocol Network Protocols FWP CSMA

22 Grant-To-Send

23 Grant-To-Send

24 Grant-To-Send C A B D

25 Grant-To-Send C A B D

26 Grant-To-Send C A B D

27 Grant-To-Send C A B D

28 Grant-To-Send C A B D

29 Grant-To-Send C A B D

30 Grant-To-Send C A B D

31 Grant-To-Send C A B D

32 Grant-To-Send C A B D

33 Grant-To-Send C A B D

34 Median Cost FWP Isolation

35 FWP Fairness Weighted Fair Queuing of protocols based on Grant-toSend durations and packet lengths.

36 Outline Survey of Failures The MNet Architecture A Design Example Further Implications

37 A Design Example: Pull Collection Protocol

38 Common Causes of Packet Loss in a Collection Protocol Disconnection Destruction Reboot Suppression Queue Egress Drop Queue Ingress Drop

39 Diagnosing Why Packets Were Lost

40 Diagnosing Why Packets Were Lost

41 Eliminating Ingress Drops Traditional Rate Control

42 Eliminating Ingress Drops Pull-Based Rate Control

43 Using FWP For Bursts

44 Using FWP For Bursts Grant of 200ms

45 Using FWP For Bursts Grant of 200ms

46 Using FWP For Bursts

47 Eliminating Ingress Drops

48 Eliminating Ingress Drops Traverse the remainder with information included in packets, used by the protocol itself

49 Outline Survey of Failures The MNet Architecture A Design Example Further Implications

50 Extensions & LimitationsThe Cost of Visibility Isolation and fairness introduce delay, increasing latencies Under light loads, we can use zero grant-tosend times.

51 Extensions & LimitationsLow Power Low-Power listening in conjunction with packet bursts.

52 Extensions & LimitationsIsolation & Fairness An individual mote must provide system isolation and fairness. Higher-layer network protocols can provide fairness across network flows

53 Comments & Questions?

54 Grant-To-Send A B C

55 Grant-To-Send A B C

56 Grant-To-Send A B C

57 Grant-To-Send A B C

58 Grant-To-Send A B C

59 Grant-To-Send A B C

60 Grant-To-Send A B C

61 Grant-To-Send A B C

62 Grant-To-Send A B C

63 Grant-To-Send A B C

64 Grant-To-Send A B C

65 FWP Fairness 150 ms 90 ms 20 ms 0 ms 50 ms Network Protocols FWP CSMA

66 FWP Isolation Median Median Cost Cost 30 1 over CSMA 1 over FWP 2 Over CSMA 2 over FWP Generation Rate (pps) 300