Efficient Network Flooding and Time Synchronization with Glossy Federico Ferrari, Marco Zimmerling, Lothar Thiele, and Olga Saukh ETH Zurich IPSN 2011 Best Paper Award Presenter: SY
Outline Introduction Design Evaluation Conclusion
Flooding Packet transmission from one node to all other Challenges Packet loss Delay Flooding storm
Glossy Flooding for wireless sensor networks Fast: 94 nodes within 2.39ms Reliable: 99.99% Scalable Time synchronization at no additional cost
Interference Capture effect Constructive interference Two signals interfere which other If one is stronger that the other Or received significantly earlier than the others Receiver might still receive the packet Constructive interference Identical packet Small Δ Δ
Generating Constructive Interference Matlab simulations
Related Works Capture effect Backcast: Dutta et al. 2008 Concurrent ACK transmission A-MAC: Dutta et al. 2010 Receiver-initiated link layer protocol
Outline Introduction Design Evaluation Conclusion
Overview Decouples flooding Concurrent transmission Constant slot length
Glossy in Detail
Timeline
Implementation Platform Challenges Tmote Sky = Taroko MSP430F1611 + CC2420 MCU and timer source by DCO temperature and voltage drifts of -0.38%/◦C and 5%/V Challenges Deterministic execution timing Start execution at same time Compensate for hardware variations
Deterministic execution timing Start reading content while receiving Immediately trigger transmission
Start execution at same time SFD interrupt Variable delay in serving interrupt Execute NOPs determined at runtime
Compensate for hardware variations Synchronizes the DCO every time Glossy starts with respect to 32.768KHz crystal Software delay uncertainty
Outline Introduction Design Evaluation Conclusion
Theoretical Analysis Scenario Worst-Case Drift of Radio Clock Assume an upper/lower bound of radio clock drift Worst-case scenario: one path at highest clock drift, another at lowest Model worst-case transmission time uncertainty Worst-case temporal displacement Uncertainty on pair of radio and MCU clock one path at minimum variation, another at maximum Worst-case temporal displacement Δ
Results Network size Node density
Controlled Experiments Setup 1 One initiator, two receivers Delay one receiver by [0,8]us Non-delay receiver@-20dBm, delayed@-13dBm
Controlled Experiments Setup 2 One initiator, variable # of recievers No delay
Controlled Experiments Setup 3 One initiator, four receivers Start a Glossy phase, computes reference time Schedules next phase All nodes activate an external pin when a phase start
Testbed Experiments Testbed Metrics Motelab: 94 nodes over three floors Twist: 92 nodes Local: 39 nodes Metrics Flooding latency L Flooding reliability R Radio on time T
Results Node density no noticeable dependency Performance depends on network size Increase N significantly enhances flooding reliability
Performance on Twist Larger size, higher latency 80% of nodes has 99.99% reliability even with lowest power Radio on time increase with network size
Maximum Number of Transmissions Vary N
Conclusion Flooding and time sync are two important services Well written, systematically analysis Promising results Detailed implementation Testbed evaluation Integrate with application might not be easy