Jayant Gupchup Phoenix, EWSN 2010 Phoenix: An Epidemic Approach to Time Reconstruction Jayant Gupchup †, Douglas Carlson †, Răzvan Musăloiu-E. †,*, Alex Szalay ±, Andreas Terzis † Department of Computer Science, Johns Hopkins University † Department of Physics and Astronomy, Johns Hopkins University ± Google *

Jayant Gupchup Phoenix, EWSN 2010 where: Environmental Monitoring

Jayant Gupchup Phoenix, EWSN 2010 Design Goals and Targets  Target Lifetime : 1 year o Duty-cycle (~ 5%)  Accuracy Requirements o Milliseconds (ms) - Seconds (s) o Online Synchronization not needed  Delay-tolerant networks o Basestation collects data opportunistically o NOT “sample-and-send”  All measurements require timestamps o Not just events

Jayant Gupchup Phoenix, EWSN 2010 Naïve Time Reconstruction  Measurements are timestamped using motes local clock  Basestation collects data  Time reconstruction algorithm: Assigns measurements a global timestamp

Jayant Gupchup Phoenix, EWSN 2010 Reconstruction is NOT Synchronization  Asynchronous operation o Each mote has its own operation schedule o No attempt to match schedules  Motes o Agnostic of network time / global time o Do not process time information o Do not have an onboard Real-Time Clock (RTC) (E.g. Telos, Mica2, MicaZ, IRIS)

Jayant Gupchup Phoenix, EWSN 2010 Phoenix Performance  Accuracy o Order of seconds, ~ 6 PPM (ignoring temperature effects)  Yield : Fraction of measurements assigned timestamps o ≥ 99%  Overheads: o Duty-Cycle: 0.2% o Space: 4%  Yield performance maintained: o Presence of random, frequent mote reboots o Absence of global clock source for months

Jayant Gupchup Phoenix, EWSN 2010 Background and Related Work

Jayant Gupchup Phoenix, EWSN 2010 Reboots and Basestation ? ? ?

Jayant Gupchup Phoenix, EWSN 2010 Cub Hill – Year long deployment

Jayant Gupchup Phoenix, EWSN 2010 Cub Hill : Time Reconstruction Nodes Stuck (Data Loss) Watchdog Fix Basestation Down Reboot Problems

Jayant Gupchup Phoenix, EWSN 2010 Rate of Reboots

Jayant Gupchup Phoenix, EWSN 2010 Reconstruction Challenges  Motes reboot at random o Downtime is non-deterministic  Dependence on basestation  Temporary network partitions  Mote clock o Varies per mote o Skew changes over time

Jayant Gupchup Phoenix, EWSN 2010 Related Work  Linear Regression for Time Rectification o Fidelity and Yield in a Volcano Monitoring Sensor Network, Werner-Allen et al., OSDI 2006  Reboot Problems o Lessons from the Hogthrob Deployments, Chang et al., WiDeploy 2008 o Trio: Enabling sustainable and scalable outdoor wireless sensor network deployments, Dutta et al., SPOTS 2006  State preservation after reboots o Surviving sensor network software faults, Chen et al., SIGOPS 2009  Data-driven Temporal Integrity o Recovering temporal integrity with data driven time synchronization, Lukac et al., IPSN 2009 o Sundial: Using sunlight to reconstruct global timestamps, Gupchup et al., EWSN 2009

Jayant Gupchup Phoenix, EWSN 2010 Phoenix

Jayant Gupchup Phoenix, EWSN 2010 Big Picture Base Station

Jayant Gupchup Phoenix, EWSN 2010 Terminology  Segment: S tate defined by a monotonically increasing local clock (LC) o Comprises  Anchor: o : Time-references between 2 segments o : Time-references between a segment and global time  Fit : Mapping between one time frame to another o Defined over : Neighbor Fit o Defined over : Global fit  Fit Parameters o Alpha (α) : Skew o Beta (β) : Offset  Goodness of Fit : Metric that estimates the quality of the fit o E.g. : Variance of the residuals

Jayant Gupchup Phoenix, EWSN Phase  Phase-I : Data Collection (In-network)  Phase-II : Timestamp Assignment (Database)

Jayant Gupchup Phoenix, EWSN 2010 Architecture Summary  Motes  Global Clock Source  Basestation

Jayant Gupchup Phoenix, EWSN 2010 Anchor Collection – I : Beaconing Each Mote: Beacons time-state periodically Beacon interval~ 30s Duty-cycle overhead: 0.075%

Jayant Gupchup Phoenix, EWSN 2010 Anchor Collection – II : Storage Each Mote: Stays up (30s) after reboot Listens for announcements Wakes up periodically (~ 6 hrs) Stays up (30s) Listens for announcements Stores anchors Duty-Cycle : 0.14%

Jayant Gupchup Phoenix, EWSN 2010 Anchor Collection – III : Global References G-Mote: Connected to a global clock source Beacon its time-state (30s) Store Global References (6 hrs) Global clock source (GPS, Basestation etc) ,

Jayant Gupchup Phoenix, EWSN (B) 43-5 (B) 97-7 (A) 97-7 (A) 28-4 (G) 28-4 (G) 97-7 (A) 97-7 (A) 43-5 (B) 43-5 (B) 28-4 (G) 28-4 (G) Time Reconstruction (outside the network) χ = 2 χ = 2.5 χ = 7 Segment Graph

Jayant Gupchup Phoenix, EWSN 2010 Evaluation: Simulation & Experiments

Jayant Gupchup Phoenix, EWSN 2010 Evaluation Metrics  Yield: Fraction of samples assigned timestamps (%)  Average PPM Error: PPM Error per measurement:  Duty Cycle Overhead: Fraction of time radio was on (%)  Space Overhead: Fraction of space used to store anchors (%)

Jayant Gupchup Phoenix, EWSN 2010 Simulation: Missing Global Clock Source Simulation Period : 1 Year

Jayant Gupchup Phoenix, EWSN 2010 Simulation: Wake Up Interval Anchor collection rate should be significantly faster than the rate of reboots

Jayant Gupchup Phoenix, EWSN 2010 Simulation: Segments to anchor with

Jayant Gupchup Phoenix, EWSN 2010 Olin Deployment - 19 Motes - 21 Day Deployment - 62 segments - One Global clock mote

Jayant Gupchup Phoenix, EWSN 2010 Deployment Accuracy

Jayant Gupchup Phoenix, EWSN 2010 Naïve Yield Vs Phoenix Yield Phoenix Yield: 99.5%

Jayant Gupchup Phoenix, EWSN 2010 Conclusion  Phoenix timestamps: o > 99% of the collected measurements o With accuracy in order of seconds  Phoenix is Robust to: o Basestation failures for days-months o Random mote reboots  Paying a price of: o 0.2% increase in duty cycle o 4% space overhead

Jayant Gupchup Phoenix, EWSN 2010 Questions ?

Jayant Gupchup Phoenix, EWSN 2010 Extra:

Jayant Gupchup Phoenix, EWSN 2010 Discussion / Future Work  Choosing the right link metric o Factor number of anchor points o Temporal separation of anchors o Combining the metrics along a “fit” path  Adaptive anchor collection o If rate of reboots is unknown  Compare with online timestamping (FTSP)

Jayant Gupchup Phoenix, EWSN 2010 Simulation Parameters ParameterTypeDefault Value Clock SkewUniform Distribution~ U (40 70) [ppm] Segment ModelNon-Parametric (Cub Hill)median : 4 days TopologyCub Hill (53 nodes) Communication Delay (end-to-end) Uniform Distribution~ U (5 15) [ms] Packet Reception RatioLog-Normal Path LossPr(2.0) = η = 2.04 σ = 6.28 Constant Constant NUM_SEGMENTSConstant4 Sampling Frequency (measurements) Constant10 mi

Jayant Gupchup Phoenix, EWSN 2010 Reboots: Long downtimes

Jayant Gupchup Phoenix, EWSN 2010 Clock Skews

Jayant Gupchup Phoenix, EWSN 2010 Temperature dependence Source: