Improving Neighbor Discovery with Slot Index Improving Neighbor - - PowerPoint PPT Presentation

▶
improving neighbor discovery with slot index improving
SMART_READER_LITE
LIVE PREVIEW

Improving Neighbor Discovery with Slot Index Improving Neighbor - - PowerPoint PPT Presentation

Aug 14, 2023 242 likes •717 views

Improving Neighbor Discovery with Slot Index Improving Neighbor Discovery with Slot Index Synchronization Synchronization Shuaizhao Jin , Zixiao Wang , Wai Kay Leong , Ben Leong , Yabo Dong , Dongming Lu Zhejiang

slide-1
SLIDE 1

1

Improving Neighbor Discovery with Slot Index Improving Neighbor Discovery with Slot Index Synchronization Synchronization

Shuaizhao Jin∗, Zixiao Wangƚ, Wai Kay Leongƚ, Ben Leongƚ , Yabo Dong∗ , Dongming Lu∗

∗ Zhejiang University, Hangzhou, China ƚ National University of Singapore, Singapore

IEEE MASS 2015 IEEE MASS 2015

slide-2
SLIDE 2

2

Docking applications Docking applications

In a docking pattern, a mobile node discovers and communicates with the static node situated at a rendezvous point. Witness Witness Search & Rescue Team Control Center Tacking of cattle movements during feeding times

slide-3
SLIDE 3

3

Tourist tracking at Mogao Grottoes Tourist tracking at Mogao Grottoes

slide-4
SLIDE 4

4

Tourist tracking at Mogao Grottoes Tourist tracking at Mogao Grottoes

More than 6,000 tourists per day!

slide-5
SLIDE 5

5

Tourist tracking at Mogao Grottoes Tourist tracking at Mogao Grottoes

Temperature : Humidity: CO2: Tourists:

slide-6
SLIDE 6

6

Challenges Challenges

Tracking accuracy No infrastructure (power supply, network) Deployment restriction

slide-7
SLIDE 7

7

Na Naï ïve tourists tracking system ve tourists tracking system

Mobile Static … … Enter Leave … Stay Cave Beacon Beacon Beacon Beacon

slide-8
SLIDE 8

8

Na Naï ïve tracking system ve tracking system

Last for nearly one year with two AA battery An average discovery latency

  • f 2.5 s

Static node 5s 30ms 5s 30ms 30ms Mobile node Daily Recharging

slide-9
SLIDE 9

9

 Improve the energy efficiency of mobile nodes

  • Duty cycling

Our goal Our goal

Static Mobile

slide-10
SLIDE 10

10

 Deterministic neighbor discovery protocols

Neighbor discovery protocols Neighbor discovery protocols

Static Mobile 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 2 Disco with a pair of primes (3, 5)

slide-11
SLIDE 11

11

 Tradeoff between duty cycle and discovery latency

Neighbor discovery protocols Neighbor discovery protocols

slide-12
SLIDE 12

12

 Latency distribution of different protocols

Neighbor discovery protocols Neighbor discovery protocols

duty cycle of 0.5% and slot size of 5ms

slide-13
SLIDE 13

13

Searchlight enough? Searchlight enough?

Statistics of tourists’ duration in caves

slide-14
SLIDE 14

14

Searchlight enough? Searchlight enough?

20% risk of discovery failure 40% measurement error 100 150

slide-15
SLIDE 15

15

 Key observation

  • Slot index offset has a significant impact on the discovery

latency

Key insight: reducing discovery latency Key insight: reducing discovery latency

Disco with a pair of primes (3, 5) Static Mobile 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 2 t1 t2 From 5 slots to 2 slots!! 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1

slide-16
SLIDE 16

16

 Average discover latency

  • Enumerate all the possible slot index offset
  • For each given slot offset, enumerate all possible contact index

Gains without slot index offset Gains without slot index offset

Duty cycle of 1%, slot size of 5ms Protocol No synchronization With index synchronization Gains Searchlight BlindDate Disco U-connect 23.56 s 31.94 s 50.63 s 55.62 s 0.33 s 0.36 s 0.32 s 0.37 s 72X 89X 158X 149X

slide-17
SLIDE 17

17

 If all the static nodes can be slot index synchronized,

the subsequent discovery latency will be largely reduced

Intuition Intuition

Mobile S1 S2 S3 Initial latency may be large! Subsequent latency will be very small!

slide-18
SLIDE 18

18

Challenges Challenges

Clock drift No direct communication between static nodes S1 S2 Range Block

slide-19
SLIDE 19

19

Key idea Key idea

S2 M S1 S3 0101…0001

slide-20
SLIDE 20

20

Absolute time synchronization? Absolute time synchronization?

Reference-based Distributed Failure of reference Reliable communication Convergence speed Closed system

slide-21
SLIDE 21

21

 MASS(Mobility-Assisted Slot index Synchronization)

  • Distributed Reference Election and Synchronization
  • Dynamic clock skew estimation and compensation
  • Mitigating the pitfalls of small synchronization errors

Our solution Our solution

slide-22
SLIDE 22

22

Distributed reference election & synchronization

slide-23
SLIDE 23

23

General idea General idea

slide-24
SLIDE 24

24

Reference election Reference election

S2 M1 S1 S3 M2 Schedule_s1 Skew_M1_s1 Schedule_s3 Skew_M2_s3

slide-25
SLIDE 25

25

Priority based election Priority based election

S2 M1 S1 S3 M2 Schedule_s1 Skew_M1_s1 Prioritys1 Schedule_s3 Skew_M2_s3 Prioritys3 Max(Prioritys1, Prioritys2, Prioritys3)

slide-26
SLIDE 26

26

 Requirements

  • Easily computed in a distributed way
  • Different for various static nodes
  • Each static node update its priority with the discovery of

mobile nodes

Node priority Node priority

slide-27
SLIDE 27

27

 Moving pattern of tourists

Node priority Node priority

slide-28
SLIDE 28

28

 Average inter-arrival time between mobile nodes

Node priority Node priority

slide-29
SLIDE 29

29

How it perform? How it perform?

Time (hr) Static node in caves Frist tourist enters Last tourist leaves Most converged in one hour Cave 29 and 12 not converge

slide-30
SLIDE 30

30

Mitigating pitfalls of small synchronization error

slide-31
SLIDE 31

31

 Small synchronization error can potentially lead to the

worst-case scenario for average discovery latency

Pitfalls of synchronization error Pitfalls of synchronization error

Striped-Searchlight(1% duty cycle)

slide-32
SLIDE 32

32

 Striped-Searchlight

Mitigating pitfalls Mitigating pitfalls

Striped-Searchlight Striped-Searchlight+1 Striped-Searchlight+1/2

slide-33
SLIDE 33

33

 Result

Gains by our modifications Gains by our modifications

slide-34
SLIDE 34

34

 Result

Gains by our modifications Gains by our modifications

Searchlight-S Searchlight-S+1 Searchlight-S+1/2 Avg Worst Avg Worst Avg Worst Latency (slots) Offset 0, 1 12.3 37 18.5 57 15.3 47 Offset 2 199.5 399 29.4 59 47.6 99 Offset 3 163.5 359 29.2 59 43.4 99 Average 125.1 265 25.7 58.3 35.4 81.7

slide-35
SLIDE 35

35

Evaluation

slide-36
SLIDE 36

36

 Setup

  • Custom trace-based simulator (C), simulating the

interactions between the mobile and static nodes

  • One-month real tourists traces collected from Mogao

Grottoes, containing 8,658 movement routes for the mobile nodes and 69,271 cave visits.

Evaluation Evaluation

slide-37
SLIDE 37

37

Effect of slot size Effect of slot size

Slot size of 5ms and 25ms under duty cycle 1%

slide-38
SLIDE 38

38

Effect of slot size Effect of slot size

Slot size of 5ms and 25ms under duty cycle 1%

55%

slide-39
SLIDE 39

39

 1% duty cycle and 5 ms slot size

Latency reduction with MASS Latency reduction with MASS

slide-40
SLIDE 40

40

 1% duty cycle and 5 ms slot size

Latency reduction with MASS Latency reduction with MASS

slide-41
SLIDE 41

41

 1% duty cycle and 5 ms slot size

Latency reduction with MASS Latency reduction with MASS

75%

slide-42
SLIDE 42

42

 Different protocols under different duty cycles

Performance gains with MASS Performance gains with MASS

Improvement with MASS with 5 ms slot size

slide-43
SLIDE 43

43

 Performance under small duty cycles

Minimum duty cycle Minimum duty cycle

slide-44
SLIDE 44

44

 Randomly produce traces based on real trace

Performance with random traces Performance with random traces

slide-45
SLIDE 45

45

 Practical deployment and experiments at

Mogao Grottoes

 Thorough study of existing neighbor

discovery protocols under the scenario of slot index synchronization

 Design more efficient protocols under small

synchronization error

Future work Future work

slide-46
SLIDE 46

46