A Software Engineering Perspective on Context-Awareness in Ad Hoc Mobile Networks Christine Julien Mobile Computing Laboratory Washington University in.

A Software Engineering Perspective on Context-Awareness in Ad Hoc Mobile Networks

Christine JulienMobile Computing Laboratory

Washington University in Saint LouisApril 18, 2023

April 18, 2023 Christine Julien 2

Motivation• Simplify mobile application development

— Individually address application-specific needs— Dynamically gather context information from the

network— Seamlessly adapt to environmental changes — Transparently help programmers cope with

problems inherent in ad hoc networks


Ad Hoc Network Challenges

• Form opportunistically • Change rapidly in response to

mobility• Rely on no fixed infrastructure• Provide transient interactions• Foster decoupled computing• Embody open environments with

limited guarantees


Computational Model• Ad hoc network• Host (mobile or stationary)

— Defines location in physical space• Agents/Application Components

— Unit of mobility residing on a host• Data

— Owned and generated by each agent

Host 1

Host 2

Host 3


Overview

• Context-aware computing• EgoSpaces conceptual model• EgoSpaces middleware• Network abstractions• Conclusions and future work


Context-Aware Computing:

The Need for Context• Software and hardware constantly move and

change• Applications must opportunistically adapt • Perception of context varies by application• Current solutions:

— Use basic types of context information in specific ways • Fieldwork tools, guide tools

— Provide access to only a single component’s sensors at a time

• Context Toolkit, Context Fabric— Support only local or symmetric interactions

• MARS, LIME



Rethinking Applications (1)

TrafficCollision







Charlie

Angie



Novel Notion of Context

• Flexible adaptation in dynamic environments— Scope that extends beyond the local host — Generalized interaction with context types

• Context needs vary by task and situation— Asymmetric context interaction— Multiple contexts evolving over time

• Context computation must be scalable— Explicit control over the scope, size, and cost


Overview

• Context-aware computing• EgoSpaces conceptual model• EgoSpaces middleware• Network abstractions• Conclusions


EgoSpaces Conceptual Model:

The View Concept• Maximal context contains all available data• A view is a projection of the maximal context

— Egocentric abstraction of operating context— Tailored to an application’s individual needs

• Allow agent to control scope of views— Ease program development— Minimize performance penalties

Host 1

Host 2

Host 3

Reference agent’s view: yellow data within one

hop



Declarative View Specification

• Allows simple yet meaningful view definitions

• Restrict context based on properties of the network, hosts, agents, and data

All location data owned by collision warning agents on cars within 100 meters of my current location

• Facilitate transparent context maintenance— Rely on underlying middleware for context

maintenance tasks

Roman, G.-C., Julien, C., and Murphy, A. L., "A Declarative Approach to Agent-Centered Context-Aware Computing in Ad Hoc Wireless Environments,” Software Engineering for Large-Scale Multi-Agent Systems, LNCS 2603, April 2003.



View Interaction

• Present view contents as tuple space• Allows use of traditional Linda-like

operations— Content-based pattern matching for tuple

retrieval• retrieve tuples (rd)• remove tuples (in)

— Can affect overlapping views— Atomic blocking and probing operationsJulien, C., and Roman, G.-C., "Egocentric Context-Aware Programming in Ad Hoc Mobile Environments," in Proceedings of FSE-10, November 2002.



Consistency Concerns

• Transactional semantics (e.g., money transfers)— Strong application guarantees— Can be expensive— Use atomic Linda-like operations

• “Best-effort” semantics (e.g., traffic)— Variety of implementations— Application chooses semantics based on its

particular situation— Use scattered probing operations



Reactive Views• Facilitate behavioral adaptation

— Agents respond to presence of certain tuples

— Associate application-level reaction to the appearance of a tuple

• Scheduling modalities correspond to consistency requirements— Eager semantics provide transactional

guarantees— Lazy semantics provide fewer guarantees

• Allow more flexible implementation• Increase performance



Active Views• Capture natural context interactions

— Transparent data migration— Automatic duplication— Event capture— Extensibility

• Simplify programming—Programming abstractions—Reduced code size

Julien, C. and Roman, G.-C., "Active Coordination in Ad Hoc Networks," to appear in Proceedings of Coordination 2004.



Access Control Requirements• Ad hoc networks are open environments, yet

agents may need to restrict data access— No permanent trusted central server

• Current work— Administrative domains

• Mobile Ambients, SEAL— Specialized type systems

• KLAIM— Fine-grained access control

• SECOS

• Must scale to large systems• Must address needs of individualized agents



Access Control Provision• Agent specified access control function

— Limits access based on agent properties— Allows fine-grained access control— Individualized to particular application’s needs

• Reference agent provides credentials and intended operations— Access control functions evaluated on a per-

tuple basis to determine inclusion in the view

• Can be used to define administrative domains


Overview



EgoSpaces Middleware:

Why Middleware?• Speed up application development in ad hoc

networks— Enhance level of programming abstraction — Provide generic and flexible context operations— Shift complex tasks to underlying infrastructure— Facilitate rapid dissemination and adoption

• Proven solutions for other domains— Distributed object systems

• CORBA, Jini, DENO— Mobile agent systems

• Aglets, D’Agents, Code— Coordination systems

• LIME, MARS, Jedi



EgoSpaces Middleware

• Programming interface simplifies application development burden— Allows declarative view specification— Presents view contents as tuple space— Provides standard Linda tuple space

operations on view definitions

• Includes light-weight implementations of tuple space operations

• Provides tailored access control



EgoSpaces Architecture

• Application provides declarative view specifications

• Middleware — Discovers neighbors — Monitors context

sensors — Provides data

contents of context to application through view abstraction

ad hoc physical network

discovery message sensing

sensor monitoring

network abstractions

EgoSpaces ELights

application



Demo Application

•Application-specific view restriction

—Hop-count based metric

—File-size restriction

•Content-based selection from view

—Songs displayed based on desired properties

•Reactive programming allows downloads to resume



Simplifying Application Development

• Abstracts network for application programmer— Removes concern for low-level network and

communication concerns— Provides high-level, abstract definition of needs

• Maintains contents of views as the environment changes— Programmer does not have to persistently poll

• Shifts developer focus to key components of applications— User interface design— Local data storage and representation


Overview



Network Abstractions:

Restricting the Network• Defining a view requires limiting

participating hosts— e.g., all hosts within five miles or

within three hops— Consider the overhead in defining

such a network— Deal with volatility of ad hoc

networks• Unannounced disconnection• Unpredictable reconnection• Heterogeneous devices

5mi

Roman, G.-C., Julien, C., and Huang, Q., "Network Abstractions for Context Aware Mobile Computing," in Proceedings of ICSE, May 2002.



Solution Overview

1. Represent the ad hoc network as a graph

2. Abstract properties of nodes and links to weights on edges in this graph

3. Calculate the cost of paths from the reference node

4. Determine the shortest path to each node and build a tree

5. Limit this tree using a bound, D



Weight Assignment• Individualized for an application• Abstracts varied properties of hosts and links

— Host properties form i; link properties form ij• i includes battery power, CPU power, location, etc.• ij includes physical distance, bandwidth, throughput, etc.

• Edge weights combine link properties and the properties of the connected nodes

mij = (i, j, ij)

ijiji

j



Computing Path Cost• Assuming each link has a single weight (mij), an

application can define a generalized cost function: v0(Pk) = Cost(v0(Pk-1), mk-1,k) v0(v0) = 0

• Cost function must be strictly increasing along a path



Limiting the Scope• Define a subtree of

the network by placing a bound on the cost of each path— Bound is useful only if

shortest path cost is strictly increasing

— Allows reference to restrict computation to a subnet

0

1

1

1

1

1

1

2

Note: All links have a weight of 1

21

2

2

3



Context Calculation Protocol• Service new queries

— Respond at the application level— Remember cost, parent, sequence number, and

information about the computation— Propagate the query

• Service shorter distances (same sequence number)— Remember cost and new parent— Propagate query

• Disregard longer distances (same sequence number)

— But remember cost and parent if within bound

• Halt when computation reaches boundary• Based on traditional ad hoc routing protocols



Protocol ExampleQuery= (source, new cost, …)

State= [cost, parent, ...]{(neighbor, cost through neighbor), …}

Bound = 6Cost = additive

c

(r, 4, …)(r, 2, …)

[0, null, …]

42

15

r

ba 1[2, r, …] [4, r, …]{(r,2)} {(r,4)}

d4

{(r,4)}{(r,2), (b,5)}

[5, b, …]{(b,5)}

[4, r, …][2, r, …]

(b, 5, …)

(b, 5, …)

(a, 3, …)

{(r,4), (a,3)}{(r,2), (b,5)}[2, r, …] [3, a, …]

[5, b, …]{(b,5)}

(b, 4, …)

{(r,4),(a,3)}{(r,2), (b,5)}[2, r, …] [3, a, …]

[4, b, …]{(b,4)}



Context Maintenance Protocol

• Persistent queries require context maintenance

• React to increase in link weight— If parent, adjust cost and propagate information— Otherwise, update local information

• React to decrease in link weight— If non-parent link, recalculate cost for neighbor on

the other end• Handle “count-to-infinity” problem

— Store entire shortest path at each hop• Regain resources when context is no longer

used



Maintenance Example

42

15

r

c

ba 1 [3, a, …]{(r,2), (b,5)} {(r,4),(a,3)}

[2, r, …]

[0, null, …]

6

[4, b, …]{(b,4)}

Query= (source, source cost, new cost, …)

State= [cost, parent, …]{(neighbor, cost through neighbor), …}

Bound = 6Cost = additive

d 4

(b, 5, …)

[5, b, …]{(b,5)}

{(r,2), (b,10)} {(r,4), (a,8)}[3, a, …][2, r, …]

{(r,2)} {(r,4)}[2, r, …] [4, r, …]



Evaluation• Simulation using ns-2 network simulator

— 100 node network in 1000x1000m2 space— Vary network density by changing

transmission range— Using random waypoint mobility model

adapted to avoid speed degradation

• Metrics— Context consistency— Settling time (at most 35ms)— Percentage of useful broadcasts



Context Building Consistency (1)

• Good consistency for reasonably sized contexts

one-hop context

two-hop context

three-hop context

four-hop context

Transmission range (meters)

Perc

enta

ge o

f co n

t ex t

rece

iving

mes

sage



Context Building Consistency (2)

• Consistency >80% with increased network load

five context definitions

one context definition

ten context definitions


Perc

enta

ge o

f co n

t ex t

rece

iving

mes

sage



Useful Broadcasts• Efficiency decreases with increasing density

one-hop context

two-hop context

three-hop context

four-hop context


Perc

enta

ge o

f us e

f ul b

road

cast

s


Conclusions

• Melding of context-awareness and ad hoc networks

• Novel notion of context— Individualized declarative specification— Includes general facets of the environment— Extended in scope to remote components

• Asymmetric context interactions• EgoSpaces model and middleware

— Supporting algorithms and protocols


Contributions (1)

• Algorithms for ad hoc mobile computing—Network abstractions—Consistent group membership

ab R

R a R

b

r

Huang, Q., Julien, C., and Roman, G.-C., "Relying on Safe Distance to Achieve Strong Partitionable Group Membership in Ad Hoc Networks," to appear in IEEE Transactions on Mobile Computing.


Contributions (2)

• Middleware for context-awareness—Addressing software engineering

concerns of mobile environments—Facilitating rapid development of

mobile applications—Bringing application development to

the hands of novice programmers

Handorean, R., Payton, J., Julien C., and Roman, G.-C., "Coordination Middleware Supporting Rapid Deployment of Ad Hoc Mobile Systems", in Proceedings of the 1st ICDCS Workshop on Mobile Computing Middleware (MCM03), May 2003.


Contributions (3)

• Formal models of context-aware and mobile computing—Specification and refinement using

Mobile UNITY—Building Context UNITY for direct

reasoning about context-aware systems•Roman, G.-G., Julien, C., and Payton, J., "A Formal Treatment of Context-

Awareness," (invited paper) to appear in Proceedings of FASE 2004.•Julien, C., Payton, J., and Roman, G.-C., "Reasoning About Context-Awareness in the Presence of Mobility," in Proceedings of FOCLASA, September 2003.

•Roman, G.-C., Julien, C., and Huang, Q., "Formal Specification and Design of Mobile Systems," in Proceedings of FMPPTA'2002, April 2002.


Future Plans

• Addressing ad hoc network and communication concerns in software engineering— Transactional programming in ad hoc networks— Tailored communication and coordination

paradigms for applications in ad hoc networks• Context-aware routing and communication

• Applying formal methods to context-aware environments— Reasoning about mobile and context-aware

applications through formal notation and proof logic


Questions?

http://www.cse.wustl.edu/[email protected]


Publications www.cse.wustl.edu/~julien

• Roman, G.-G., Julien, C., and Payton, J., "A Formal Treatment of Context-Awareness," (invited paper) to appear in Proceedings of Fundamental Approaches to Software Engineering 2004.

• Julien, C. and Roman, G.-C., "Active Coordination in Ad Hoc Networks," to appear in Proceedings of Coordination 2004.

• Huang, Q., Julien, C., and Roman, G.-C., "Relying on Safe Distance to Achieve Strong Partitionable Group Membership in Ad Hoc Networks," to appear in IEEE Transactions on Mobile Computing.

• Julien, C., Payton, J., and Roman, G.-C., "Reasoning About Context-Awareness in the Presence of Mobility," in Proceedings of the 2nd International Workshop on Foundations of Coordination Languages and Software Architectures (FOCLASA03), September 2003.

• Handorean, R., Payton, J., Julien C., and Roman, G.-C., "Coordination Middleware Supporting Rapid Deployment of Ad Hoc Mobile Systems", in Proceedings of the First International ICDCS Workshop on Mobile Computing Middleware (MCM03), May 2003.

• Roman, G.-C., Julien, C., and Murphy, A. L., "A Declarative Approach to Agent-Centered Context-Aware Computing in Ad Hoc Wireless Environments (extended version)," Software Engineering for Large-Scale Multi-Agent Systems, A. Garcia et. al. (editors), LNCS 2603, April 2003.

• Julien, C., and Roman, G.-C., "Egocentric Context-Aware Programming in Ad Hoc Mobile Environments," in Proceedings of FSE-10, November 2002, pp. 21-30.

• Roman, G.-C., Julien, C., and Huang, Q., "Network Abstractions for Context Aware Mobile Computing," in Proceedings of the 24th International Conference on Software Engineering (ICSE'02), Orlando, FL (USA), May 2002, pp. 363-373.

• Roman, G.-C., Julien, C., and Murphy, A. L., "A Declarative Approach to Agent-Centered Context-Aware Computing in Ad Hoc Wireless Environments," in the 1st International Workshop on Software Engineering for Large-Scale Multi-Agent Systems (SELMAS'2002), May 2002.

• Roman, G.-C., Julien, C., and Huang, Q., "Formal Specification and Design of Mobile Systems," in Proceedings of FMPPTA'2002, April 2002.

A Software Engineering Perspective on Context-Awareness in Ad Hoc Mobile Networks Christine Julien Mobile Computing Laboratory Washington University in.

Documents

context types context

context software

perception of context

time context toolkit

time context computation

context fabric support

view concept maximal

christine julien3