building efficient wireless networks with low-level naming csci 599 ies spring 2002: building...

37
CSCI 599 IES Spring 2002 : Building Efficient Building Efficient Wireless Networks Wireless Networks with with Low-Level Naming Low-Level Naming Presented By: - Jayman Dalal Jayman Dalal [email protected]

Post on 21-Dec-2015

215 views

Category:

Documents


0 download

TRANSCRIPT

CSCI 599 IES Spring 2002:

Building Efficient Building Efficient Wireless Networks with Wireless Networks with

Low-Level NamingLow-Level Naming Presented By: -

Jayman DalalJayman [email protected]

CSCI 599 - IES Spring 2002, USC

AuthorsThe authors are researchers at USC/Information Sciences Institute. Work carried out as a part of SCADDS project.

• John Heidemann

• Ramesh Govindan

• Deborah Estrin

• Fabio Silva

• Chermek Intangonwiwat

• Deepak Ganesan

CSCI 599 - IES Spring 2002, USC

Presentation Outline• Motivation

• Introduction & Related Work

• Architecture

• Implementations

• Application Techniques

• Experiment Results & Evaluation

• Summary and Critique

CSCI 599 - IES Spring 2002, USC

Motivation

• Emergence of Sensor Networks

• Lack of a concrete software architecture supporting following in sensor networks:– named data

– in-network processing

– nested queries

• Unavailability of operational testbed

CSCI 599 - IES Spring 2002, USC

Introduction & Related Work

• Comparison of conventional and sensor networks.

• Key contributions of the research work

• Related attribute based naming systems

• Related systems using in-network processing

• Comparable sensor-network-specific systems

CSCI 599 - IES Spring 2002, USC

Comparison of Conventional & Sensor Networks

Conventional Networks

• High Bandwidth

• Low Delay

• Throughput is primary constraint

Sensor Networks

• Plentiful processing time

• Bandwidth primary constraint

CSCI 599 - IES Spring 2002, USC

Key Contributions of this Research Work

• Attribute-based naming scheme with flexible matching rules.

• Showing how this approach to naming enables application specific, in-network processing

CSCI 599 - IES Spring 2002, USC

Related Attribute based naming systems• Attribute based naming systems based on top of general

purpose networks

– Univers and Yellow-pages naming at UA

– X.500 and LDAP

• Attribute based communication for structuring distributed systems

– Structuring distributed programs using CPUs

– ISIS and Information Bus

• Using attribute based primitives for specific problems

– Reliable Multicast communication using named data

– Distributed Whiteboard

CSCI 599 - IES Spring 2002, USC

Related Systems using In-Network Processing

• Active Services

– Assumes roughly equivalent distances between all service providing nodes

– This work assumes communication cost between nodes vary greatly

• Active Networks

– Target internet-like domains

– This work targets sensor network domains

• Adaptive Web Caching and Peer-to-Peer file processing

CSCI 599 - IES Spring 2002, USC

Comparable Sensor-Network-Specific Systems• Internet Ad-Hoc Routing

– Does not support in-network processing

• Jini & Ninja Service Discovery Service– Distribute processing to user nodes– Suitable for LAN – High bandwidth

• Piconet– Tiered architecture and energy conserving– Lacks attribute based naming and in-network processing

• SPIN– Globally unique identifiers for individual sensors– Does not consider application-specific in-network processing

CSCI 599 - IES Spring 2002, USC

Comparable Sensor-Network-Specific Systems• Intentional Naming Systems

– Very similar in use of attributes and dynamically location of devices

– Uses an overlay network over IP-based Internet

• LEACH – Use of in-network data compression

• DataSpace– Use of IPv6 multicast addresses corresponding to geographic

locations.

• COUGAR– Use of centralized query translation and distributed processing

• Declarative Routing– Very similar to this research work

CSCI 599 - IES Spring 2002, USC

Architecture

Communication Architecture Components

• Directed Diffusion

• Attribute Tuples and Matching Rules

• Filters

CSCI 599 - IES Spring 2002, USC

Directed Diffusion

• A data-centric communication paradigm for sensor networks

• Goal is to establish an efficient n-way communication mechanism

CSCI 599 - IES Spring 2002, USC

Directed Diffusion - Definitions

• Interest: A list of attribute-value pairs describing a task

• Sink: Node originating the interest

• Source: Sensor node that matches the interest, collects data and sends it back

• Gradient: Direction towards which data matching and interest flows and status of demand

CSCI 599 - IES Spring 2002, USC

Directed Diffusion

Source

Event

Interests

Sink

Gradients

CSCI 599 - IES Spring 2002, USC

Directed Diffusion

• Intermediary data is cached as it propagates from source to sink.

• Reinforcement

• Negative Reinforcement

CSCI 599 - IES Spring 2002, USC

Attribute Tuples & Matching Rules

• Interests and data messages are composed of attribute-value-operation tuples.

• Keys: Identify attributes. Drawn from central authority.

• Operations: Define interactions of data messages and interests. e.g. GT, LT, GE, LE, EQ, EQ_ANY, IS.

• Actual: Literal or bound value

• Formal: Comparison or unbound value.

CSCI 599 - IES Spring 2002, USC

Attribute Tuples & Matching Rules

• One-way Match: Compares all formal parameters of one attribute set against actuals of another.

given two attribute sets A and Bfor each attribute a in A where a.op is a formal {

matched = falsefor each attribute b in B where a.key =

b.key and b.op is an actualif a.val compares with b.val using a.op, then matched = true

if not matched then return false(no match)}

• Complete Match: One-way match succeeds in both directions

CSCI 599 - IES Spring 2002, USC

Interactions of Diffusion & Matching• Consider an example user query

{class IS interesttype IS four-legged-animal-searchinterval IS 20msduration IS 10sx GE -100x LE 200y GE 100y LE 400}

• Sensor detects something it responds with{class IS datatype IS four-legged-animal-searchinstance IS elephantx IS 125y IS 220intensity IS 0.6confidence IS 0.85timestamp IS 1:20}

CSCI 599 - IES Spring 2002, USC

Filters• A novel concept.

• Application-specific code invoked when data enters a node.

• Influences manipulation of data, caching of data and its further route.

• Can be hard coded at design time or distributed as mobile code packages.

CSCI 599 - IES Spring 2002, USC

Implementations

• SCADDS diffusion version 3 or the macro diffusion (PC/104 Node)

• MIT Lincoln Labs’ Declarative Routing (WINSng 1.0 node)

• Micro Diffusion (UCB Rene Mote

CSCI 599 - IES Spring 2002, USC

Macro Diffusion & Declarative Routing

• Run on Linux on desktop PCs and PC/104-based sensor nodes and on WINSng 1.0 sensor nodes.

• Use of publish/subscribe APIs having event-driver programming style.

• Differences:– Filters

– Routing Mechanism

CSCI 599 - IES Spring 2002, USC

Micro Diffusion

• Scaled down version containing gradients, single tag attributes and limited filters

• Runs on 8 bit CPU and 8KB memory

• Motes and micro diffusion can be used in areas requiring dense sensor distribution.

CSCI 599 - IES Spring 2002, USC

Application Techniques for Sensor Networks

• In-network data aggregation

• Nested Queries

CSCI 599 - IES Spring 2002, USC

In-network Data Aggregation

• Multiple detection of target results in unnecessary communications.

• Energy can be conserved if data is aggregated:

– Binary value – there was a detection.

– An area – there was a detection in quadrant 2.

– An application specific aggregation.

• Caching of data by intermediate sensors. Suppresses propagation of duplicate data.

CSCI 599 - IES Spring 2002, USC

Nested Queries

• Triggering of secondary sensor based on status on primary.

• Advantage: Triggered sensor directly interprets the initial sensor’s data => Reduction in network traffic and latency

• Challenges:– How to robustly match initial

and triggered sensors?

– How to select a good triggered sensor?

Simple Approach

Nested Approach

CSCI 599 - IES Spring 2002, USC

Evaluation

• Aggregation Benefits

• Nested query Benefits

CSCI 599 - IES Spring 2002, USC

Node positions in sensor testbed at ISI

CSCI 599 - IES Spring 2002, USC

Aggregation Benefits

• Sink – 28; Source – 13, 16, 22, 25. 4 Hops.

• Measures aggregate bytes sent by diffusion for networks with and without data aggregation.

• Suppression is able to reduce network traffic by 42% for four sources.

CSCI 599 - IES Spring 2002, USC

Nested Query Benefits• User – 39; Audio Sensor –

20; Light Sensors – 13, 16, 22, 25

• One hop from light sensor to audio, and two hops from there to user.

• Measures percentage of light change events that successfully result in audio data delivered to user.

• Flat queries suffer more loss than nested queries

CSCI 599 - IES Spring 2002, USC

Summary

• An approach to communication in highly constrained distributed systems like sensor networks focusing on attribute-based naming and in-network processing.

• In-network processing with filters, data aggregation, nested queries help reduce network traffic and conserve energy.

CSCI 599 - IES Spring 2002, USC

Critique – Main Contribution

• Introduces a topology independent attribute-based naming system for low-level communication.

• Uses application-level in-network processing using filters, data aggregation and nested queries.

• This technique provides many advantages for wireless sensor networks by reducing communication overhead and overall energy consumption.

CSCI 599 - IES Spring 2002, USC

Critique – Claims

• Topologically independent low-level naming helps in reducing transmission thereby conserving energy

• In-network processing done using nested queries and filters, close to the place where data is sensed, reduces communication costs.

CSCI 599 - IES Spring 2002, USC

Critique – Assumptions

• Sufficient amount of storage space is available to cache the amount of data generated.

• In-network processing time is minimal and does not affect real-time communication, if required.

CSCI 599 - IES Spring 2002, USC

Critique – Concerns

• No clear-cut definition of “a neighbor”.

• Security Issues: Replacing the node with a malicious one or signal trapping.

• Needs to have more robust communications.

• Issues regarding mobility of nodes. Does the gradients still remain valid?

CSCI 599 - IES Spring 2002, USC

Questions

CSCI 599 - IES Spring 2002, USC