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

Improving Neighbor Discovery with Slot Index Improving Neighbor Discovery with Slot Index Synchronization Synchronization Shuaizhao Jin ∗ , Zixiao Wang ƚ , Wai Kay Leong ƚ , Ben Leong ƚ , Yabo Dong ∗ , Dongming Lu ∗ ∗ Zhejiang University, Hangzhou, China ƚ National University of Singapore, Singapore IEEE MASS 2015 IEEE MASS 2015 1

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 Control Center Search & Rescue Team Tacking of cattle movements during feeding times 2

Tourist tracking at Mogao Grottoes Tourist tracking at Mogao Grottoes 3

Tourist tracking at Mogao Grottoes Tourist tracking at Mogao Grottoes More than 6,000 tourists per day! 4

Tourist tracking at Mogao Grottoes Tourist tracking at Mogao Grottoes Temperature : Humidity: CO 2 : Tourists: 5

Challenges Challenges No infrastructure (power supply, network) Deployment restriction Tracking accuracy 6

Naï ïve tourists tracking system ve tourists tracking system Na Enter Stay Leave Cave … … … Static Beacon Beacon Beacon Beacon Mobile 7

Naï ïve tracking system ve tracking system Na 5s 5s Static node 30ms 30ms 30ms An average discovery latency Last for nearly one year with of 2.5 s two AA battery Mobile node Daily Recharging 8

Our goal Our goal  Improve the energy efficiency of mobile nodes  Duty cycling Static Mobile 9

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

Neighbor discovery protocols Neighbor discovery protocols  Tradeoff between duty cycle and discovery latency 11

Neighbor discovery protocols Neighbor discovery protocols  Latency distribution of different protocols duty cycle of 0.5% and slot size of 5ms 12

Searchlight enough? Searchlight enough? Statistics of tourists’ duration in 13 caves

Searchlight enough? Searchlight enough? 100 150 20% risk of discovery failure 40% measurement error 14

Key insight: reducing discovery latency Key insight: reducing discovery latency  Key observation  Slot index offset has a significant impact on the discovery latency Disco with a pair of primes (3, 5) 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 2 Static 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 Mobile From 5 slots to 2 slots!! t2 t1 15

Gains without slot index offset Gains without slot index offset  Average discover latency  Enumerate all the possible slot index offset  For each given slot offset, enumerate all possible contact index Duty cycle of 1%, slot size of 5ms Protocol No synchronization With index synchronization Gains Searchlight 23.56 s 0.33 s 72X 31.94 s 89X BlindDate 0.36 s Disco 50.63 s 0.32 s 158X 55.62 s 149X U-connect 0.37 s 16

Intuition Intuition  If all the static nodes can be slot index synchronized, the subsequent discovery latency will be largely reduced S1 S2 S3 Mobile Subsequent latency Initial latency may will be very small! be large! 17

Challenges Challenges No direct communication Range between static nodes S2 S1 Block Clock drift 18

Key idea Key idea 0101…0001 S3 S2 S1 M 19

Absolute time synchronization? Absolute time synchronization? Failure of reference Reference-based Reliable communication Convergence speed Distributed Closed system 20

Our solution Our solution  MASS(Mobility-Assisted Slot index Synchronization)  Distributed Reference Election and Synchronization  Dynamic clock skew estimation and compensation  Mitigating the pitfalls of small synchronization errors 21

Distributed reference election & synchronization 22

General idea General idea 23

Reference election Reference election Schedule_s1 Schedule_s3 Skew_M1_s1 Skew_M2_s3 S3 S2 S1 M1 M2 24

Priority based election Priority based election Schedule_s1 Schedule_s3 Skew_M1_s1 Skew_M2_s3 Priority s1 Priority s3 S3 S2 S1 Max(Priority s1 , Priority s2 , Priority s3 ) M2 M1 25

Node priority Node priority  Requirements  Easily computed in a distributed way  Different for various static nodes  Each static node update its priority with the discovery of mobile nodes 26

Node priority Node priority  Moving pattern of tourists 27

Node priority Node priority  Average inter-arrival time between mobile nodes 28

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

Mitigating pitfalls of small synchronization error 30

Pitfalls of synchronization error Pitfalls of synchronization error  Small synchronization error can potentially lead to the worst-case scenario for average discovery latency Striped-Searchlight(1% duty cycle) 31

Mitigating pitfalls Mitigating pitfalls  Striped-Searchlight Striped-Searchlight Striped-Searchlight+1 Striped-Searchlight+1/2 32

Gains by our modifications Gains by our modifications  Result 33

Gains by our modifications Gains by our modifications  Result Searchlight-S Searchlight-S+1 Searchlight-S+1/2 Latency (slots) Avg Worst Avg Worst Avg Worst 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 34

Evaluation 35

Evaluation Evaluation  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. 36

Effect of slot size Effect of slot size Slot size of 5ms and 25ms under duty cycle 1% 37

Effect of slot size Effect of slot size 55% Slot size of 5ms and 25ms under duty cycle 1% 38

Latency reduction with MASS Latency reduction with MASS  1% duty cycle and 5 ms slot size 39

Latency reduction with MASS Latency reduction with MASS  1% duty cycle and 5 ms slot size 40

Latency reduction with MASS Latency reduction with MASS  1% duty cycle and 5 ms slot size 75% 41

Performance gains with MASS Performance gains with MASS  Different protocols under different duty cycles Improvement with MASS with 5 ms slot size 42

Minimum duty cycle Minimum duty cycle  Performance under small duty cycles 43

Performance with random traces Performance with random traces  Randomly produce traces based on real trace 44

Future work Future work  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 45

46