DIFS: A Distributed Index for Features in Sensor Networks Benjamin Greenstein, Deborah Estrin, Ramesh Govindan, Sylvia Ratnasamy, and Scott Shenker Department of omputer Science, University of California at Los. 2003.11.06 指導教授 : 陳朝鈞 報告者:李衍儀 2016/7/13 1 Contents • Introduction • Approach – GHT: A Geographic Hash Table/ Structured Replication – Quad Tree – DIFS • Conclusion 2016/7/13 2 Introduction • • Energy constrained sensor network. Local Storage. Small number of queries and more collected data. Store event locally and query to all nodes. Directed Diffusion, TAG: aggregation and pruning for statistics. Inherent cost of flooding. (flood-then-respond) • Data-centric storage (DCS). Storing data by name. GHT/ Structured Replication. Not efficient for range query. Challenge: How to support range query efficiently? 2016/7/13 3 GHT/ Structured Replication • Key/value-based distributed index Event : Key. Hashed to geographic position • Hotspot Problem: Structured Replication • Structured Replication Make 4d-1 images. Store closest mirror. Query to all mirror nodes. Keys are an event names. Range queries over the value is not efficient 2016/7/13 4 Quad Tree • Indexing values for range queries. Spatially distributed quad tree of histogram. • Efficient range queries. Branches can be pruned during search. • But, not scalable Every data information must be propagated to the root. Every queries must originate with root. 2016/7/13 query 5 DIFS (1/5) • Multiply rooted hierarchical index. Nodes can have multiple parents. Nodes store information for a particular range of values detected within a particular geographic region. • Load balance Higher-level nodes cover smaller value ranges within large region. Lower-level nodes cover larger value ranges within small region. • Efficiently support range queries, queries about distribution of value in space. 2016/7/13 6 DIFS (2/5) 6 bfact: 2i • • 2016/7/13 Each child have bfact parents. The range of child is bfact times the range of parent. 7 A Simplified Example [0…3]; Full-N/W [0…7]; Q1 [0…7]; Q2 [4…7]; Full-N/W [0…7]; Q3 [0…7];Q4 bfact = 2 Max Levels = 2 Max Range [0…7] 2016/7/13 8 Conclusion • DIFS: index structure for sensornets for energy-efficient and load balanced range search. • DIFS performs best when Many queries posed relative to the number of generated events. The result sets is small. The result sets is not likely to be distributed the network. • DIFS is scalable to a large number of searches or stores. 2016/7/13 9