Envirostore: A Cooperative Storage System for Disconnected

Download Report

Transcript Envirostore: A Cooperative Storage System for Disconnected

Infocom’07
Authors:Liqian Luo, Chengdu Huang, Tarek Abdelzaher
John Stankovic
Presented By Rohini Kurkal
Under Guidance of Dr.Bin Tang
Contents
 Introduction
 System Design
 Implementation
 Performance Evaluation
 Possible Extensions
Introduction
Applications:
-Environmental monitoring/data logging
 No real time communication b/w sensor and sink
 Disconnected system
EnviroStore - cooperative storage system for sensor
networks suitable for disconnected operation
It is storage-centric rather than communication-centric
Major concern -maximizing effective storage capacity
Implemented in nesC for TinyOS and evaluated in TOSSIM
Storage centric paradigm
 Should be simple and lightweight
 Micaz has only an 8MHz 8-bit processor and a 4KB RAM,
 Iris has 8KB RAM, 512KB Measurement Flash Memory
 Sensor nodes do not maintain files but just writes data to
collection station and never read the data they write
Data redistribution–must improve overall storage
utilization
System Design
Sink – process that runs on a user’s PC, identified by a
regular IP address and a well known TCP port
Data mule-collects data wirelessly from encountered nodes
and dumps these data later to the base station
 Two types of data mules:
- Intentionally relay data b/w the sink and the sensor nodes
- Opportunistic data upload
System Model
System Design
Data redistribution is used to maximize the effective storage
space of the sensor network
 In-network Data Redistribution:
 Sensor nodes are in a single network
 Overloaded nodes offload data to neighboring empty nodes
 Cross-partition Data Redistribution:
 Disconnected network
 Overloaded nodes upload to data mules
 Data Mules offload to under-loaded nodes
In-network data redistribution
 Uses lazy-offload scheme to save energy – postpones data
balancing until the storage overflows
 Overloaded nodes should satisfy below conditions:
Ri < RTH and
Ri’ - Ri > Rimbalance
Where ,
Ri = Remaining Storage size
RTH = Threshold to delay data transfer
Rimbalance = Parameter to allow local imbalances
Ri’ = Average remaining storage
Contd..
 Bad Idea:
 Selecting neighbor with largest remaining free space
 This can cause data ping-pong
 Prevent data ping pong - bound the amount of data
transfer
 Remaining storage & remaining node energy must be
checked
 Node should not invoke or accept data redistribution
unless its estimated energy lifetime > estimated storage
lifetime
Cross-partition data redistribution
 Uses data mules
 Discriminate nodes (overloaded and under-loaded):
 Data mule calculates its free storage value R’m as the weighted
sum αR’+(1-α)Rm.
α = 1 : mule favors redistribution to neighborhood
α =0 : emphasizes upload
Conserve power and reduce message collision, nodes use
back-off timers
Transition state of sensor node
System Architecture of sensor nodes
Implementation
 Implemented using nesC in Tinyos
 Local storage space of nodes is organized into a circular
buffers
 Uses:
 It consumes minimum code and data memory
 organizes space as continuous data chunk
 eliminates the need for free space management mechanisms
 prolongs flash lifetime by balancing write access to different
locations
Log Files
Log-array file :
Simultaneously written by different nodes
Generates a sequence of log items
Logs attributes of environmental events independently
monitored by multiple nodes
Log-sequence file:
one writer at a time
 Multiple nodes must coordinate with each other
Maintains unique & continuous serial numbers
Used in EnviroSuite
Log-Array File
Log-Sequence File
PerformanceEvaluation
 Basic deployment configuration: field of 80 × 80 ft2 ,
36 nodes were deployed
Two Scenarios:
Scenarios 1: Single Disconnected
sensor network
Scenario 2 :Partitioned
sensor network with data mules
Comparison of data storing rate at different time
Why EnviroStore is different ?
 Used for disconnected sensor networks
 Extra constraint of limited energy – use lazy offload
 Resource limitation of individual nodes
 Load balancing must be dependent only on local
information
 Has additional challenge of redistributing data
between entities that are disconnected
Possible Extensions
 use of controllable data mules to optimize data
redistribution and upload
 data replacement policies to maximize the total
amount of information instead of just the amount of
stored data
 performance evaluation of EnviroStore on real
hardware platform
THANK YOU