A Distributed ib t Sleep Scheduling Protocol under Loosely Synchronized Clocks for Wireless Networks David (Bong Jun) Choi (Supervisor: Xuemin (Sherman) Shen) Broadband Communications Research Lab. Department of Electrical and Computer Engineering University of Waterloo
Scenario 1: Conference Conditions Each hour you choose to be either at (1) the conference hotel or (2) your room Objective Two persons (you and Professor) want to meet each other With the least number of visits to the conference hotel Knowledge They agreed to meet at 3:00 PM What would you do? Attend the conference at 3:00 PM Scheduled (Synchronous) 2/18
Scenario 2: Conference Conditions Each hour you choose to be either at (1) the conference hotel or (2) your room Objective Two persons (you and Professor) want to meet each other With ihthe least number of visits ii to the conference hotel Knowledge Each person knows that the other person will be at the conference today But, do not know when? What would you do? Visit the conference hotel several times during the day? How often? When? Unscheduled (Asynchronous) 3/18
Scenario 3: Conference Conditions Each hhour you choose to be either at (1) the conference hotel or (2) your room Objective Two persons (you and Professor) want to meet each other With the least number of visits to the conference hotel Knowledge Each person knows that the other person will likely be at the conference today around the lunch buffet What would you do? Probabilistic? bili Somewhat/Loosely scheduled (Semi-Asynchronous) 4/18
Research Outline DSA: Distributed Semi-Asynchronous Sleep Scheduling for Mobile Wireless Networks What? Design a sleep scheduling protocol for wireless mobile devices in a distributed network topology considering loosely synchronized clocks Why? How? Reduce energy consumption Practical to assume some level of synchronization accuracy Using slotted frame structure with predetermined allocation of active slots Allocation depending on the synchronization error estimation 5
Overview Introduction Synchronization and Energy Saving Semi-Asynchronous Sleep Scheduling System Model Proposed Protocol Performance Evaluation Conclusion 6
Introduction: Synchronization Timekeeping which requires the coordination of events to operate a system in unison Applications 1) WSN: Accurate timing Information for data collection 2) TDMA: Slot alignment 3) PSM: Sleep Scheduling 7
Introduction: Timekeeping in Wireless Communications NTP (Network Time Protocol) GPS (Global Positioning System) a) )Quartz crystal clocks b) Chip scale atomic clock by NIST 8
Introduction: In reality, Unavailability of GPS / Intermittently available energy, cost, isolation Imperfect clocks Inherent oscillator error External factors: temperature, pressure, noise, aging, etc. Error increase with time and number of hops [TTP] MCXO: Microcomputer-compensated tdcrystal tloscillator illt TCXO: Temperature-compensated crystal oscillator 9 [TTP] K. Romer, Time Synchronization in Ad Hoc Networks, in Proc. ACM MobiHoc, 2001. B.M. Sadler. "Fundamentals of energy-constrained sensor network systems, IEEE Aerospace and Electronic Systems Magazine, Tutorial Supplement, 2005.
Introduction Synchronization Error & Energy Efficiency Sleep scheduling requires accurate timing for coordinating awake/active periods power beacon period power guard periods node i X node i clock offset clock offset node j X node j beacon/search interval time beacon/search interval time Larger Longer Higher Synchronization Awake Periods Energy Error (Guard Periods) Consumption 10
System Model Clock Model: C(t) t) ock time, C( clo fast clock C (t)> 1 perfect clock C (t)= 1 slow clock C (t)< 1 0.6 ~ 6.0 ms error in 60.0 s UTC 11
Synchronous vs. Asynchronous vs. Semi- Asynchronous Scenario 1: power slot (= frame) node i 0 0 Scenario 2: node j power slot 0 0 search interval a) Synchronous: node i... 1 2 3 4 5 6 7 1 2 3 4 5 6 7... t node j... 1 2 3 4 5 6 7 1 2 3 4 5 6 7... frame (= search interval) t 12/18 b) Asynchronous:
Synchronous vs. Asynchronous vs. Semi- Asynchronous Scenario 3: power slot guard interval node i -2-1 0 1 2-2 -1 0 1 2 node j -2-1 0 1 2-2 -1 0 1 2 frame synchronization error search interval t c) Semi-asynchronous: 13/18
System Model Sleep Scheduling Model Frame Structure One Slot per Frame! Existing synchronous protocols only use slot 0 GAPS! Existing asynchronous protocols does NOT work n gi L s C i (t 1 ) Node i -3-2 -1 0 1 2 3-3 -2-1 0 1 2 3 t L f C ij (t 1 ) T SI Node j -3-2 -1 0 1 2 3-3 -2-1 0 1 2 3 n gj Search Interval (T SI ) Determined by [AEB, STAR] SI C j (t 1 ) T SI : search/probing interval L F : frame length L S : slot length C ij : clock offset between nodes i and j t 14 [STAR] W. Wang, V. Srinivasan, and M. Motani, Adaptive Contact Probing Mechanisms for Delay Tolerant Applications, in Proc. ACM/IEEE MOBICOM, 2007. [AEB] B. J. Choi and X. Shen, Adaptive Exponential Beacon Period Protocol for Power Saving in Delay Tolerant Networks, in Proc. IEEE ICC, 2009.
Scenario 4: Individual Differences Conference Conditions Each hour you choose to be either at (1) the conference hotel or (2) your room Objective Two persons (you and Professor) want to meet each other With the least number of visits to the conference hotel Knowledge Each person knows that the other person will likely be at the conference today around the lunch buffet Professor loves to be at the conference hotel, but you don t What would you do? Probabilistic? Somewhat/Loosely scheduled (Semi-Asynchronous) Individual differences in the Time Window 15/18
System Model Synchronization Error Model [TTP, RBS, TPSN] Sync error increase with time Sync error decrease by means of synchronization algorithm L f changes with time L s L g Node i -3-2 -1 0 1 2 3-4 -3-2 -1 0 1 2 3 4 L f t 16 [RBS] J. Elson, L. Girod, and D. Estrin, "Fine-Grained Network Time Synchronization Using Reference Broadcasts," SIGOPS Oper. Syst. Rev., vol. 36, no. SI, pp. 147 163, 2002. [TPSN] S. Ganeriwal, R. Kumar, and M. B. Srivastava, "Timing-Sync Protocol for Sensor Networks," in Proc. ACM SenSys, 2003.
Problem Definition Need to guarantee overlapping of at least one active slot between two frames with time shifts Semi-Asynchronous: L f < T SI Bounded Error Bounded Shift: n max = max /L s cf.) Synchronous (L f = 1 slot ), Asynchronous (L f = T SI ) 17
Problem Definition An Illustrative Example n g = 3 E 3,3-3 -2-1 0 1 2 3 S(E 3,3, 0) E 4,3-4 -3-2 -1 0 1 2 3 4-4 -3-2 -1 0 1 2 3 4 S(E 4,3, -7) S(E 4,3, -6)... -4-3 -2-1 0 1 2 3 4 S(E 4,3, 0) n g = 4-4 -3-2 -1 0 1 2 3 4-4 -3-2 -1 0 1 2 3 4... S(E 4,3, +6) S(E 4,3, +7) t 18/18
Proposed Protocol CGI: Traditional Method (= Continuous) n: number of slots for The estimated sync error DSA: Basic Idea - Frame E n consists of c consecutive active slots (for all nodes) Followed by active slots that are c c slots apart (for individual error) An active slot on the last slot The beacon sent every c slots (not just on slot 0 as in CGI) 19
Numerical Result Comparison of Active Ratio (Vs. Asynchronous & CGI) 1.0 s, < 0.4 s Larger error (n max ) larger active ratio Outperforming region: large sync error to T SI ratio 20
Simulation Result Simulation Setting ONE Simulator [ONE] + Sync Algorithm + Sleep Scheduling Protocol Mobile Nodes, Sparse Density 21 [ONE] A. Keranen, J. Ott, and T. Karkkainen, "The ONE Simulator for DTN Protocol Evaluation," in Proc. ICST SIMUTools, 2009. [18] G. Anastasi, A. Falchi, A. Passarella, M. Conti, and E. Gregori, "Performance Measurements of Motes Sensor Networks," in Proc. ACM MSWiM, 2004.
Simulation Result Simulation Scenarios 1. Reference Node based Synchronization [DTP] At least one time server Client nodes tune to the reference time in the server Sync Error: increase 100 us/s 2. Rl Relative Clock ksynchronization i [SA, AD] No time server Nodes cooperatively tries to minimize relative time error Averaging between contacted nodes Sync Error: Initial configuration (< 100 ppm) a) Reference node based b) Distributed / Relative 22 [DTP] Q. Ye and L. Cheng, "DTP: Double-Pairwise Time Protocol for Disruption Tolerant Networks," in Proc. IEEE ICDCS, 2008. [SA] M. Sasabe and T. Takine, "A Simple Scheme for Relative Time Synchronization in Delay Tolerant MANETs," in Proc. International Conference on Intelligent Networking and Collaborative Systems, 2009. [AD] Q. Li and D. Rus, "Global Clock Synchronization in Sensor Networks," IEEE Trans. Computers, vol. 55, no. 2, pp. 214 226, 2006.
Simulation Result Reference Node Based Lower speed Slower Convergence Speed Larger error Higher Energy Consumption EC of DSA: 20% @ 1.0 m/s, 30% @ 5.0 m/s 23
Simulation Result (Distributed) Relative Synchronization Lower speed Slower Convergence Speed Larger error Higher Energy Consumption Converge (assuming control parameters are all deterministic) EC of DSA < EC of fcg CGI (especially for large errors, fixed c) 24
Conclusion Clock Synchronization & Energy Consumption Practical to assume some level of clock accuracy with error bounds Distributed Semi-Asynchronous Sleep Scheduling Protocol (DSA) Periodic Active Slots + Minimum Active Slots (not shown) )Optimized i for given distribution ib i of synchronization i error Performance Evaluation Comparison of Active Ratio vs. Existing Protocols Simulation Results Further Work Overhead from using synchronization protocols for loosely synchronized clocks vs. Asynchronous protocols Adaptive version of DSA for time varying distribution of sync error Extensive simulations 25
Adaptive DSA? Some problems in the Relative synchronization simulation results! DSA is already adaptive to the time varying synchronization error Min c is fixed! c can be adjusted dusing the distribution ib i of the long term synchronization error 26/18
Thank You Questions & Comments? David (Bong Jun) Choi bjchoi@engmail.uwaterloo.ca