Storage Issues in Sensor Networks

1
Storage Issues in Sensor Networks

• Peter Desnoyers
• Vijay Sundaram
• David Yates
• University of Massachusetts Computer Science

2
The Role of Storage in Sensing Networks

• Sensors produce data continuously

Applications use data in batches
3
Storage Bridges the Gap

• Storage adapted to the needs of sensing networks
  is needed to
• Receive data when it is generated
• Find data when it is needed
• Privide data to multiple applications at varying
  times

4
SensorNet Storage Research at UMass
Self-managing Storage
Sensor Field
Data collection, archival, and processing
Smart Caching
Distributed Storage
5
Distributed Storage for Sensor Networks
Peter Desnoyers

• Network traffic is costly or slow in sensor
  networks
• Local storage is fast and cheap
• So use it! But how?

Problem develop efficient methods to organize,
index, and retrieve data stored at local nodes in
sensor systems
6
The WriteHere Storage System
Target
Take advantage of data locality to minimize data
collection and indexing overhead
Construct efficient height-balanced distributed
index trees

• Current work
• optimized index algorithm to exploit data
  locality in sensor systems
• prototype system to demonstrate applications
  using distributed storage

7
Self-Managing Storage
Vijay Sundaram
Storage
Multiple Applications
Disk performance varies with placement and access
pattern As much as 1,000,000 to 1
Problem distribute and schedule disk reads and
writes to meet per-application performance
guarantees
8
Self-Management Techniques
Bandwidth Allocation
Automated Placement within systems
Requests
Associate Request with Application Class
Convey New Allocation
QoS aware disk scheduler
Bandwidth Reallocation Module
Convey response time
Monitor response times
Dynamic Placement across systems

• Current research
• hotspot detection
• cost-aware rebalancing

• Measure, detect, rebalance
• NP-hard problem
• Use intelligent heuristics

9
Data Caching and Dissemination
David Yates

• Applications have different cost and quality
  constraints

• Sensing network bandwidth is costly
• External network bandwidth is cheap
• Use a cache

Problem exploit characteristics of sensor data
to implement smart caching
10
Smart Caching Strategies

• Prefetch data
• Shift queries in time or space
• Deny low priority queries when busy
• Just get the data

• Current work
• Determine relative benefit of strategies in
  simulation
• Validate and quantify benefits in NETRAD testbed