SCADDS: Research Update October 2000 Deborah Estrin, Ramesh Govindan, John Heidemann USC/ISI and UCLA SCADDS Staff and Students: Jeremy Elson, Deepak Ganesan, Chalermek Intanagonwiwat, Fabio Silva, Jerry Zhao For more information: http:/www.isi.edu/scadds Research Update – Directed diffusion studies • Update • Aggregation • Multipath – Systems contributions • API and implementation for Diffusion and SenseIT routing • Address free fragmentation – Experimental platform and experience • PC-104s • Instrumentation/debug support! – Plans and related projects • Aggregation and multipath simulations and implementations • Adaptive fidelity evaluations • Related projects: Localization, Time synchronization, Tags, Tiered architecture PART I: Algorithm/Protocol/Diffusion Studies • Diffusion recap • Aggregation • Multipath SENSIT PI-MTG October 00 3 Diffusion-Recap • Directed diffusion (Joules/Node/Received Event) Average Dissipated Energy 0.03 0.025 – Can provide significantly longer network lifetimes than existing schemes – Keys to achieving this: Diffusion without suppression 0.02 0.015 flooding 0.01 Omniscient multicast 0.005 Diffusion with suppression 0 0 50 100 150 200 250 300 Network Size (nodes) SENSIT PI-MTG October 00 • In-network aggregation • Empirical adaptation to path 4 Latency in Data Diffusion Compare latency with: • flooding: large amount of traffic causes delay • omniscient multicast: theoretical centralized optimum (unrealizable in practice) • data diffusion without suppression • data diffusion with suppression Delay (Seconds) 0.8 0.7 0.6 Diffusion without suppression 0.5 0.4 0.3 0.2 flooding 0.1 0 Diffusion w/suppression 0 50 100 150 o. multicast 200 250 Network Size (nodes) 300 Diffusion’s empirical adaptation and in-network processing (suppression) achieves latency as low as optimum (o. multicast). 5 Diffusion Status • Preliminary simulation results were presented in Mobicom 2000 (and April00 PI meeting) • Diffusion version 1 integrated into current ns snapshot and released to research community • A simple TDMA MAC is implemented in ns for better simulations of sensor radio – Tracking other researchers group TDMA work for future incorporation (e.g., Srivastava et. al.) SENSIT PI-MTG October 00 6 Diffusion Work in Progress • Aggregation mechanisms for energy savings • Multipath SENSIT PI-MTG October 00 7 Aggregation Dissipated Energy 0.03 0.025 Diffusion- No suppression 0.02 0.015 Flooding Omnicient Multicast 0.01 0.005 0 • Application-level data processing can improve energy efficiency Diffusion 0 50 100 150 200 250 300 • Opportunistic and greedy aggregation • Distributed aggregation points automatically and locally selected such that they are close to sources • Opportunistic: aggregation on existing tree • Greedy: use reinforcement to increase aggregation closer to sources..favoring energy reduction over latency 8 Simplified Problem Statement • Where should network aggregate ? Data Source 1 B C A New Data Source 2 D E Sink F – B, C, D, E, or F? • If aggregation reduces size only slightly – F is acceptable, “shortest path tree” – “opportunistic aggregation” minimizes latency to sink • If aggregation reduces size significantly – D is preferred (closer to A), “greedy(ier) tree” – Conserved energy compared to F9 – May increase A to F latency Simplified Problem (Continued) Data Source 1 • Naïve local-rules may not work B – If local rule always favors C A New Data Source 2 aggregated data paths, B may be selected as aggregation point— inefficient and higher latency D E Sink F SENSIT PI-MTG October 00 10 Desired Aggregation Behavior [x1,y1,SNR1] B [x2,y2,SNR2] A C Sink Gradient Low rate data Reinforcement • A sample local reinforcement rule to provide “greedy(ier)” tree – A, already getting source [x1,y1] data at high rate from neighbor B – A receives [x2,y2] aggregatable data from neighbor C – A decides whether to aggregate at A or let B (upstream neighbor) aggregate – if (DelayViaB-DelayViaC < d), A reinforces B, else reinforces C - d is an adjustable parameter11 Desired Aggregation Behavior [x1,y1,SNR1] B [x2,y2,SNR2] A C Sink Gradient Low rate data Reinforcement • A sample local reinforcement rule for new data [x2, y2, SNR2] – if A sees ( delay(B)-delay(C) < d) then A reinforces B, else reinforces C – B is an upstream neighbor that has a high-rate gradient toward A for data that is aggregatable with new data [x2, y2, SNR2] - d is an adjustable parameter SENSIT PI-MTG October 00 12 Challenges • Some aggregation/processing problems are more challenging than others • Future work: – “Bounding box” applications as initial target – More general applications will require additional mechanism • identify classes of problems for which opportunistic aggregation does not produce imprecise or incorrect results • establish error bounds for class of problems for which opportunistic aggregation produces imprecise results SENSIT PI-MTG October 00 13 Multipath for Low-Latency Robustness in Lossy Networks • In the same design space as FEC and spread spectrum approaches to minimize losses and latency due to disturbances in the network • Use local rules for redundancy in lossy regions to achieve higher likelihood of delivery. • Local metrics for Path selection – Latency – Loss – Energy Shaded regions correspond to regions of high losses. Darker shades correspond to greater losses SENSIT PI-MTG October 00 14 Braided Multipath • Disjoint Paths – Stringent restriction – Allow end-to-end decisions only – Unsuitable for broadcast model Braided multi-path • Braided paths – enable distributed decision making – Offers greater flexibility to route around losses – May offer greater robustness for same energy constraints – May be better suited for changing losses in the network. Alternate path (higher latency) 15 Exploring Multipath • Exploring tradeoff between choosing higher latency path that avoids regions of high losses vs sending redundant packets through lossy regions • Exploring Localized mechanisms for low-energy notifications – Piggybacking on data packets – Nodes use notifications to trigger multipath explorations • Tradeoff-increased latency SENSIT PI-MTG October 00 16 Adaptive Fidelity • extend system lifetime while maintaining accuracy • approach: – estimate node density needed for desired quality – automatically adapt to variations in current density due to uneven deployment or node failure – assumes dense initial deployment or additional node deployment SENSIT PI-MTG October 00 zzz zzz zzz zzz 17 Adaptive Fidelity Status • applications: – maintain consistent latency or bandwidth in multihop communication – maintain consistent sensor vigilance • status: – probablistic neighborhood estimation for ad hoc routing • 30-55% longer lifetime with 2-6sec higher initial delay – currently underway: location-aware neighborhood estimation SENSIT PI-MTG October 00 18 Part II: System Developments • API for Diffusion/Network Routing • Using Random Identifiers SENSIT PI-MTG October 00 19 Integration Participation • Coordinated integration effort – BAE (Signal Processing) – ISI-W (Diffusion Routing) – Penn State (CSP) • Included 4 SensIT nodes along the road – Local detection of vehicles – Messages exchanged via Diffusion SENSIT PI-MTG October 00 20 Diffusion Routing Implementation • Two implementations: – WinCE (WINS NG 1.0 Nodes) – PC104s + Radiometrix Radios or Wired • • • • • Main development platform Easily portable to QNX Develop various in-house applications Evaluate implementation Gain experience with API SENSIT PI-MTG October 00 21 Diffusion Routing API • Objective: Improve current Network Routing API to better match distributed applications needs • Solution: Allow more control over routing decisions and packet forwarding – Support in-network processing and aggregation with flexible application interface SENSIT PI-MTG October 00 App 1 App 2 Diffusion 22 Future Directions • TDMA • Release updated network routing API after gaining experience with inhouse experiments SENSIT PI-MTG October 00 23 Random Transaction Identifiers • Maximize usefulness of every bit – each bit transmitted reduces net lifetime – can’t amortize large headers or claim-collide overhead for low data rates + high dynamics • Still need to identify transmitter – Reinforcements, Fragmentation • Use small, random transaction identifiers (locally selected…like multicast addresses) – Treat identifier collisions as any other loss • Address-free method wins in networks with locality – simultaneous transactions at any one point is much less than in network as a whole Example: A model of address-free fragmentation (16 bit data) AFF Allows us to optimize # bits used for identifiers Fewer bits = fewer wasted bits per data bit, but high collision rate; vs. More bits = less waste due to ID collisions but many bits wasted on headers SENSIT PI-MTG October 00 25 Testbed Validation of AFF Collision Model: 5 Transmitters and 1 Receiver SENSIT PI-MTG October 00 26 Part III: Experimental Infrastructure SENSIT PI-MTG October 00 27 Platform for experimentation with SCADDS algorithms • Complementary platform to Sensoria nodes: – Not for desert-field testing ! COTS, rather than custom lowpower, real-time, integrated sensor platform • Can provide larger scale networking studies and flexibility via COTS • Model: explore on this testbed and feedback lessons to integrated, Sensoria platform • Will be much easier to move back and forth with any Unix variant (e.g., QNX) • Specifications: – COTS PC104 CPU module • AMD ELANSC400, 16MB RAM+16MB FlashDisk, 4 serial/1 parallel ports – Radio: 418Mhz RPC from Radiometrix • Moving to RFM – OS: Slimmed Redhat 6.1. (2.2.x/Libc6) SENSIT PI-MTG October 00 28 Using Testbed for SCADDS Experimentation • Expanded the testbed size to explore SCADDS related algorithms – Currently 30, Target 50-100 • Debugging/Management Utilities – Special debug-stations with Ethernet and 8-serialport adapters, acting as a bridge for interactive debugging from host PCs. – CVS-like Scripts to automatically update binaries when newer version is available. • Iteratively improving SCADDS algorithms based on experimental feedback – E.g., per-hop filters underway since v.1 – Validating and feeding back into simulation results SENSIT PI-MTG October 00 29 Leveraging Tiered architecture • Leveraging other funding to enrich SCADDS experiments • Designing “Tags” under a complementary NSF grant (NSF SCOWR and ONR DURIP) – Modular architecture, reusable components • Module Bus: 80pin connector: I2C, INTQ/A and GPIOs • Modules: PIC based master module, sensor module, RFM based radio module. – Experiments with low power architecture • Software selectable clocking – Also collaborate with UC Berkeley folks to incorporate their silver-dollar –sized “motes”. • Developing a beaconing application to complement SCADDS testbed as well as an objecting tracking application. SENSIT PI-MTG October 00 30 *Photo From http://www.cs.berkeley.edu/~jhill/ Planned Work • Diffusion – – – – Aggregation simulation and implementation Multipath simulation and implementation Exploring power-aware and geographic routing assist Adaptive fidelity • Testbed experimentation • Beyond SCADDS – Timing and coordinate synchronization – Localization (ranging and self-configuring beacon placement) – Sensor network health monitoring and debugging Other collaborators: Nirupama Bulusu, Alberto Cerpa, Lewis Girod, Satish Kumar, Yan Yu SENSIT PI-MTG October 00 31