Reliable, Robust Data Collection in Sensor Networks Murali Rangan Russell Sears Fall 2005 – Sensornet

Motivation Data collection apps like habitat monitoring  All of the logged data is valuable  Real time / in order delivery not important  In-network aggregation not required Current data collection mechanisms  Lose data during temporary network partitions  Or, store all data in each node Our goal  Reliable, robust scheme to collect ALL the data in a steady stream, without in order guarantees

Our approach Reliability  Link level, hop-by-hop acknowledgements  Introduce redundancy with k replicas of data  Replicas ensure their data reaches the base station, or pass responsibility to a parent. Take advantage of broadcast medium Take advantage of existing routing algorithms (like MintRoute) for parent selection Keep basic mechanism simple, add optimization on top

Algorithm overview Node with new data  Broadcasts request for k volunteers (for replicas)  Parent(s) agree to volunteer  Broadcasts data to volunteer(s)  Volunteer(s) broadcast hop ACK (HACK)  Once k volunteers are found, deletes data Volunteer transfers responsibility upstream  Repeats above process, seeking 1 volunteer  Once responsibility is transferred, deletes data Can volunteer for same data again!

Status and Future Work Basic algorithm details are fleshed out Implement and simulate using TOSSIM  Work in progress! Address the following challenges  How many replicas? Depends upon network density, reliability  Base station broadcasts ACK to remove replicas of collected data Use bloom filters to summarize? Limit broadcast to ‘interested’ nodes?  How to bound k replicas? Eavesdropping?  How do we deal with fan-in near the base station? Few parents / replicas share paths

Prerequisites Use any routing protocol that  Assigns ‘parent(s)’ to each mote  Adapts to changes in network topology Reliable link level delivery  Make use of broadcast medium  Transfer data from a child node to its parents  ACK’s transfer responsibility for a piece of data. (Dropped ACK’s lead to more replicas) Broadcast ACK from base station  Guarantees replicas are eventually deleted  Efficient, reliable broadcast? Bloom filters? Reverse path routing + eavesdropping?

Sketch of algorithm Inserting data into the network  Broadcast request for k volunteers  Broadcast data, volunteers (parents) ACK  Once k volunteers are found, node deletes data. Volunteers  request enough volunteers to create k replicas in network. (Usually, just create 1 volunteer)  Pass ACK’s to children after creating > 1 replica. Challenges  How many replicas?  What about fan-in? Replicas share < k parents Shared paths?  How can we optimize propagation?