Transcript Sensor Networks Storage

Sensor Networks Storage
Sanket Totala
Sudarshan Jagannathan
Outline





Introduction
Storage Mechanisms
Multi-Resolution Storage
Two-Tier Storage
Conclusion
Introduction
 Sensor Nodes




Low Memory
Low Power
Low processing capability
Inapproachability
 Sensor Networks
 Efficiency requirements
 Highly dense
Introduction
 Sensor Storage





Large number of events
Query handling capability required
Streaming data
Aging mechanism
Data organization
Sensor Network Storage Mechanism
 Centralized Storage
 Central server
 Ample power
 Sufficient storage
 Single point of failure
 Fast Query processing
 High communication
 Sparse Networks
 Infrequent events
 Low Data Transfer
 Scalability problems
Sensor Network Storage Mechanism
 Distributed Storage
Storage at each node
Local computation
Scalable
Dense Networks
 Frequent events
 High Data Transfer
 Slower Query processing
 Flooding
 Distributed Indexing
 Drill-down querying




Multi-Resolution Storage
 Multi-Resolution summarization
 Construct data summaries
 Hierarchy construction
 Drill-down query evaluation
 Narrowing search space
 Spatio-temporal compression
 Data aging
 Efficient storage utilization
 Data Degradation
Multi-Resolution summarization
 Temporal Summarization
 Exploits Temporal
redundancy
 Computation overhead
 No communication
overhead
 Spatial Summarization
 Exploits Spatial redundancy
 Data Summarization
every level
 Hierarchy construction
Drill-down query evaluation
 Spatial compression
 Finer data view with
every level.
 Reduced search cost
 Query routing
 Sub region selection
 More accurate result
Data Aging
 Long deployment
 Limited storage
 Efficient resource
utilization
 Fast query processing
 Accuracy
 Higher level, Higher
time periods
Data Aging – Algorithms
 Omniscient Algorithm
 Data sets available
 Full global knowledge required
 Query specific
 Training Algorithm
 Data sets available
 Data set partitioned
 Training data (available during sensor deployment)
 Test data
 Greedy Algorithm
 Data set unavailable
 Assigns weights to data summaries
Outline – Two Tier Sensor Storage
 Design Considerations and Principles
 System Design
 Architecture
 Data structures
 Data Storage
 Sensor Network Data Summarization
 Conclusion
Design Consideration and Principles
 The Three-Tier Model
 Bottom tier Untethered sensor
nodes
 Middle tier - Tethered
sensor proxies
 Upper tier Applications and user
terminals
Design Consideration and Principles
 Principles
 Store locally, access globally
 Distinguish data from metadata
 Provide data-centric query support
Data Structures Used
 Skip Graph
 Ordered index
 In-place
indexing
 Log n height
 Probabilistic
balance
 Redundancy
and resiliency
Data Structures Used
 Interval Skip Graph




Extends skip graphs to store intervals
Allows efficient searches
Complexity of search is O(log n)
Insertion cost of O(n)
 Sparse Interval Skip
Graph
Data Storage
 Local Storage at Sensors




Archival Stores: Collection of records
Interval skip graphs used
Efficient routing and query handling
Operations: Create, Read & Delete
Sensor Data Summarization
 Data summaries –
bind the storage at the
remote sensor and the
index at the proxy
 Each update from a
sensor to the proxy
includes
 The summary
 Time period
corresponding to the
summary
 The start and end
offsets for the flash
archive
Sensor Summarization
 Adaptive Summarization
 Balances the cost of sending updates
against the cost of false positives
 Summarization Parameters
 The interval over which summaries of the
data are constructed and transmitted to the
proxy
 The size of the application specific
summary.
Conclusion
 Multi-tier nature of sensor networks
 Large amounts of events and data
 Lossy Approach
 Two-Tiered Approach
 Low overheads
 Decentralized and Hierarchical storage
 Possible Solution
 Combine lossy nature of Multi-resolution with
routing techniques of Two-tiered approach.