Large Scale Semantic Service Discovery in the Future Internet

1

• Large Scale Semantic Service Discovery in the
  Future Internet
• April 2007

2
Contents

• Introduction
• Well known protocols
• Service discovery in ad-hoc networks
• Semantic service discovery
• Large scale semantic service discovery
• Context aware service discovery

3
Why service discovery?

• Human will be surrounded by a various computing
  devices.
• Tiny sensors, PDA, PC, CP, notebook, server
• Extreme complexity to manage those devices
• Zero-administration, Zero-configuration
• Need to facilitate interaction between the
  computer
• gt Goal of Service discovery
• Originally, to lower the burden of system
  configuration
• Plug and play or zero configuration
• In more dynamic or ad-hoc environments, service
  discovery is a necessity.

4
Service Discovery

• What is service discovery?
• A protocol which enables users to discover the
  most appropriate services to the given context by
  automatically detecting the services available in
  the network.
• Components and issues
• Directory repository
• Directory structurization
• Service description and matching
• Semantic representation and matching
• Query and service announcement
• Semantic routing

5
Why service discovery?

• Scenario 1
• Mr. Sue visits ICU
• He searches the Web and finds an on-line Map
  using his PDA.
• But its too small to view on his PDA
• PDA locates the printers
• Mr. Sue (or system) picks up a closest printer
  among the public printers that are allowed to be
  used by the guests
• PDA requests printing service (without having a
  driver for the printer) without Mr. Sues
  intervention
• We need to find appropriate services
• Printing service, Wireless connection service and
  Location information service.
• Service discovery system will do that with your
  minimum intervention

6
Pervasive Service Discovery vs Web service
discovery

• Web Service Discovery
• No physical location limitation
• Focuses only on interoperation among applications
• Interoperability through standards such as WSDL
  and XML
• Universal Description, Discovery, and
  Integration(UDDI)
• The discovery and configuration process
  analysts, programmers, administrators
• Registry and data structure mainly for EC. too
  specific for Pervasive computing service
• Pervasive Service Discovery
• Ambient Services Discovery Local
• Focuses also on both among applications and
  people
• Integration with people and their ambient
  environments.

7
Pervasive Service Discovery

• Integration with People
• Integration of computing devices with people
• How do we protect personal privacy?
• Personal information persons presence, even
  health status from wearable medical device,
  users intention..
• How much prior knowledge a user or service
  provider must have for service discovery?
• Ambiguity print service, printing service,
  standard service name?
• Integration with Environments
• How can we precisely define the ambient
  environment ?
• Location, current user tasks,
• Visitors view differs from hosts view
• Heterogeneity
• H/W, S/W platform, network protocols common
  platform?
• Dynamic conditions
• Time based approaches soft state and leased
  based approaches

8
Service Discovery

• Research Trend

Context-aware Discovery - Context-based Ranking
Pervasiveness
Semantic Discovery -Semantic representation
Matching
Discovery in Ad-hoc Networks - Mobility,
Minimizing Cost
Discovery in Large-scale Networks - Structured
architecture (e.g. DHT)
Discovery in LAN -Jini, UPnP, SLP, Salutation
Static Directory Service -X.500, LDAP
Time
9
Service Discovery

• Standards / Specifications
• Discovery Protocols
• Jini, UPnP, SLP, Salutation
• UDDI
• Web Service Dynamic Discovery (WS-Discovery)
• complementary to UDDI which is focusing on
  dealing with devices and systems that are not
  always connected to the network (by MS, Intel,
  Canon )
• WSMX (Web Services Execution Environment)
• an execution environment which enables discovery,
  selection, mediation, invocation and
  interoperation of the Semantic Web Services (WSMX
  working group)
• Includes WSMO and WSML
• Service Description
• Web Service Description Language (WSDL)
• SOAP Service Description Language (SSDL)
• Web Service Modeling Language (WSML)
• Ontology architecture for Semantic Description
• DAML-S, OWL-S
• Web Services Modeling Ontology (WSMO)

10
Service Discovery

• Challenges
• Existing researches have considered the
  functionality and performance issues in isolation
• They should be considered at the same time

Service Discovery
require
require
Rich Functionality
Good Performance
Semantics-based matching
Heterogeneous networks
Scalability
Context-based ranking
Scalable directory management and message
forwarding
Need for adaptation to the characteristics of
underlying network
The best service can be different according to
the given context
Users need not to have pre-knowledge about
services to specify the desired services
11
Components

• Announcement
• Register
• Multicast/broadcast
• Query/Service Access
• Syntax
• Ontology
• Programming language dependency

• Directory repository
• Centralized
• Distributed
• Hierarchical
• Structured P2P
• Ad hoc
• Service description
• Attribute/value
• Tree-like
• XML
• Ontology (DAML, OWL)

12
Service and attribute naming

• Service Naming
• printing, print
• Naming standard to avoid the naming conflict
• Bluetooth maps service names to 128-bit number
• SLP service template define format for name
  attribute
• Apples Rendezvous (Bonjour) based on Internets
  DNS
• When a mobile client moves to another service
  directory domain, it need to understand different
  service protocol and vocabularies.
• Solution?
• How to support new services?
• Client should know new service name and then
  learn the new term.
• Its easy to add new service name to a service
  protocol standard.
• Attribute naming
• Standard naming convention
• More precise query requirements, fewer selected
  services
• Wild card searches
• Let clients examine all the available services

13
DAML/OWL Language

• OWL Web Ontology Language
• a revision of the DAMLOIL web ontology language
  
• Extends vocabulary of
• XML and RDF/S
• Rich ontology representation language
• Language features chosen for efficient
  implementations

Web Languages RDF/S XML
DAML-ONT
DAMLOIL OWL
OIL
Formal Foundations Description Logics
Frame Systems
FACT, CLASSIC, DLP,
14
Directory Design

• Centralized directories
• It stores all the services information
• Bottleneck, SPF, inefficient in large service
  discovery system
• Distributed directories
• Store services information within their own
  domain
• A service lookup may go through several
  directories.
• Number of copies
• Single copy
• Multiple copies Jini, SLP
• More reliable
• INS fully replicated copy only go to the
  directory where a client attached
• Consistency
• Flat vs. Hierarchical Directory structure
• DNS
• Difficult to make directories both scalable and
  efficient

15
Directory Design

• Service state in Directories
• Soft state most protocol
• Will not be valid after expiration (service
  announcement include life span)
• No need to monitor service state even when
  service is down.
• need regular service announcement
• Hard state
• Keep until it is told to change
• Few service announcements and housekeeping jobs
• Difficult to guarantee that all service
  information is up to date
• Service will go down without notification
  network communication error
• Solution? polling
• Directory address
• DNS listen on well known ports
• Moving to different networks directory address?
  If no DHCP?
• Multicast address Directory listen to and talk
  on a multicast address SLP, SSDS
• Multiple response? client pick one. Good for
  fault tolerance

16
Service selection

• User vs. Protocol Selection
• To find services for users efficiently and
  accurately
• Protocol selection
• Little user involvement
• No reflect the actual users will
• User selection
• Tedious for a user to examine many printers and
  compare them
• Balance?
• Best match
• Matches the best service based on application
  defined metrics INS
• Context-awareness
• Location, connection, persons situation
• Scope-awareness
• To support a large amount of services, defining
  and grouping services in scope location(INS with
  Cricket), optional attribute for location (Jini)
• Administrative domain information multiple
  hierarchy directory
• Yellow pages, White pages service information
  in non-leaf directory hashed (SSDS)
• QoS-awareness
• Better service less loaded services or better
  resource price ration services
• Most protocols only support static attribute
  not load of the printer

17
Well-known protocols

• Jini
• UPnP
• SLP
• Bluetooth SDP
• Salutation

18
Jini - introduction

• Java-based and distributed network-connecting
  technology by Sun Microsystems
• Network Plug-And-Work
• Enable spontaneous networking
• Promote service-based architecture
• A federation of clients and services
• Entities in federation provide and/or obtain
  services to/from other entities
• All developments in Java
• Code mobility
• Relies heavily on
• Object serialization
• RMI Remote Method Invocation

19
UPnP

• Service advertisement and discovery architecture
  supported by UPnP Forum
• Peer-to-Peer Model
• Based on current Internet protocols and
  technologies
• XML/HTTP, RPC
• No mobile codeinstead, standardized protocols
  and service descriptions
• XML-based service descriptions
• UPnP V1.0 Spec in 2000

20
UPnP Six Steps

• Addressing ? AutoIP
• For the IP address impaired
• Discovery ? SSDP
• Discovery and advertisement
• multicast discovery support
• XML UDP unicast and multicast packets to
  advertise their services
• Description ? (HTTP)
• What are the characteristics of a service?
• Control ? SOAP
• Making a service do its thing(s)
• Eventing ? GENA
• Updates on interesting service state changes
• Presentation ? (HTTP)
• Device GUI

21
Service Location Protocol (SLP)

• IETF standard protocol for service discovery and
  advertisement
• Designed solely for IP-based networks
• Doesnt define the protocol used between the
  client and server

• Standard Case

22
Salutation Introduction

• The Salutation Architecture was invented to
• Solve the problems of service discovery and
  utilization
• Among broad set of appliances and equipments
• In an environment of widespread connectivity and
  mobility
• Find and Bind on a network
• Devices
• Applications
• Services
• Open Architecture
• Independent of
• Operating System
• Network Protocol
• Product Class

23
Service discoveries in ad-hoc networks

• No central directory service
• Limited use of network/device resources
• Distributed
• Main entities
• Node acts as client server
• Service announcement
• Flooding How to reduce flooding traffic?
• Periodically, delta announcement (Konark)
• Slotted (DEAPSpace uses slottedperiodic)
• Cache advertisements
• TTL

24
Service discovery in ad hoc networks

• Service description
• What is described?
• service name, type, attribute, keywords,
  properties and functions
• Service hierarchy - tree like structure (Konark,
  GSD)
• Interface format, e.g., function prototype.
• How is it described and stored?
• WSDL file (Konark)
• Data structures similar to ASN.1 (DEAPSpace)
• Encoder/decoder
• Ontology OWL, DAMLOIL (GSD)
• Access to the service
• RPC
• SOAP/HTTP (Konark)
• Specific Interface, e.g., function prototype
• Encoder/decoder (DEAPSpace)
• Query formation
• Path based or syntax based
• Request routing
• Policy based (Allia)

25
Service discovery in ad hoc network
26
Semantic service discovery

• Difficulties of Service Discovery
• Different Resources
• Computing Devices, Software Services, Information
  Sources
• Representations, Capabilities, Usage
• Distinct Environmental Characteristics
• Preference, Permission, Context
• Research Issues
• Abstract Representations
• Semantic Matching
• Context-awareness
• Use of semantic ontology
• Inexact/exact querying
• More powerful reasoning engines and AI tools
• Enhancing current service location protocols

Find me the best Pizza service
27
Context vs. Semantics

• Context
• Any information that can be used to characterize
  the situation of entities (i.e. whether a person,
  place or object)
• Identity, Location, Status (or Activity), Time
• Semantics
• Abstract notions which can be implicitly derived
  to identify the differences and correlations
  between objects/concepts
• What is the semantics of Alarm?
• Definition in dictionary
• Sudden fear produced by awareness of danger
• A noise warning of danger
• A bell and etc. which sounds to warn of danger or
  to wake a person from sleep

28
Semantic service discovery

• DReggie 8
• Enhanced Jini lookup service to enables semantic
  service discovery of Jini enabled services.
• Target application M-commerce
• Two main concepts for service discovery
• DAML based service ontology
• service components and their some properties such
  as service name, capability, requirements.
• Prolog reasoner for semantic discovery
• Enhanced SDP in Bluetooth 10
• Semantic service discovery in ad hoc network
• Two main concepts for semantic matching
• DAML based service ontology
• Prolog-based XSB Reasoning engine
• Limitation
• Does not assume pervasive environment need
  components for evaluation of dynamic attributes
• Ontology restriction describe only service

29
Location-aware service discovery

• Where is the closest Italian restaurant to me?
• Location sensing
• Nomadic users
• Handy devices

30
CoolTown

• Every service (e.g., printer) has a web server,
  and a tag.
• Users walk in and receive URL from tag beacons
• Connect to the URL which describes the service
  and can invokes it
• Place manager maintains resources in a place,
  with web interface (HTML and XML)
• Resources can be grouped.
• Also acts as a resolver for some places

31
Splendor

• Tag-based location aware
• Tag label location and people
• Proxies
• Enhanced security
• Service management (registration, security)

Trusted servers managing services
Bootstrapping using multicast
32
Agents2Go
PalmApp (with CDPD modem)
Cell tower
Location identified by cell tower id
register
Restaurant agent or other services
Initial query form
User query
Brokers
Area 2
User query
Locator
Area 3
Maps geographic area to broker
Agents2Go Server
33
Contributions

• Attach web URL to services and periodically
  broadcast (CoolTown)
• Intermediate trusted proxy for security and
  service management (Splendor)
• Localized brokers store services in the location
  and are mapped to based on the user location
  (Agent2Go)

34
Internet-scale discovery

• Construct scalable, robust services
• Huge throughput demands and availability
• Interoperability
• Efficiency
• Fault tolerance

INS/Twine, Ninja/SDS, WSPDS
35
INS/Twine

• Uses Chord
• structured peer to peer system
• Lookup in Log(N)
• Service Propagation
• Resource state changes, update is propagated
• Periodic refreshing of resource information
• edge resolvers attached to clients update
  frequently
• Internal resolvers infrequently

root
subject
res
resolver
camera
traffic
model
man
AModel
ACompany
Resource is a camera, manufactured by ACompany
and filming traffic.
StrandMapper Hash(strand)
Distributed hash table (Chord). KeyRouter maps to
Chord.
30 INS/Twine a scalable p2p architecture for
intentional resource discovery, 2002
36
Ninja

• Secure Service Discovery Service (SDS)
• Uses XML to encode service descriptions
• Certificate Authority and Capability Managers for
  security
• Servers dynamically organize themselves into
  hierarchies for scalability
• Filters query upwards to avoid overloading one
  node (Bloom filter)

Services running on clusters of workstations
Distributed data structures
A chain across proxies and bases to find better
service
Stateless and soft-state proxy
Heterogeneous devices and sensors
37
Multi-tier ubiquitous service discovery protocol
Global Services Registry-gateway to global network

• Context attributes provided by services
• Distance to server
• Server load
• Service channel
• Attribute-value pair service description
• Localized Service Propagation
• propagate upward with attenuation factor of
  service attributes
• Query Routing
• Overlay hierarchy

Brokering Agent-recommends appropriate service
Global Services
26 A multi-tier ubiquitous service discovery
protocol, Mobisys, 2003
Domain Hierarchy
38
Ontology based semantic service discovery (ICU 03)

• Existing Service Discovery
• Automatic discovery of network devices and
  services
• Automatic configuration and reconfiguration of
  network devices and services
• Limitation
• Lack of service representation such as their
  functionality and capability
• Service description does not include semantic
  meaning
• Syntactic level service discovery only support
  exact matching
• Attribute-based matching Jini, UPnP, Salutation
• Object-based matching RMI registry, CORBA
  naming service, UDDI
• Ontology based semantic service discovery is
  needed

39
Introduction challenges

• Design and development of efficient and
  specialized ontologies for a pervasive home
  environment
• Rich description of devices and services
• To make possible various service request message
  type
• Represents all kinds of device and service in
  home network environment.
• device attributes, device status, service state
  variable, control interface, and their
  relationship
• Extensibility
• New device and service should be added without
  modifying ontology.
• Develop service discovery system to determine
  whether the developed ontologies can be
  applicable
• Most suitable service should be discovered in
  real-time

40
Ontologies in home environment

• Relationship among device, service and primitive
  service

Audio Device
Embedded device- Mike
Sound primitive service
Alarm service
Clock primitive service
Sound primitive service Clock primitive service
Player primitive service
Embedded device- Speaker
Recorder primitive service
Radio primitive service
41
Implementation

• Architecture of Extended Lookup Service

42
Activity Policy-based Service Discovery Framework
for Pervasive Computing(ICU 06)

• Semantic Service Discovery (Gaia, Aura, MIDAS,
  ABSD, TCE)
• Abstract representations
• Semantic matching
• Use of domain-specific ontologies for context
  sharing and reuse
• Context-aware Service Discovery (Anatomy, Solar,
  CAttr, CaOb)
• Exact or abstract representations
• Syntactic or semantic matching
• Use a variety of contexts
• gt not consider dynamic changes of service
  semantics in different contexts

43
Problem Statement

• Because the semantics of services is dynamically
  changed in different contexts, the acquired
  services might be inappropriate in some
  situations
• For Example
• Abstraction maximizes the availability of
  services
• MusicPlayer gt WinAmp or WindowsMediaPlayer
  (from Gaia)
• EditText gt MSWord or Emacs (from Aura)
• Context-awareness minimizes user distractions
• Near MusicPlayer gt ServiceTypeMusicPlayer,
  Location getCurrentLocation(Woo-Hyun)
• Available EditText gt ServiceTypeEditText,
  ExecutionEnvgetCurrentExecutionEnv()
• The changes of semantics in different contexts
  make both of the benefits useless
• Bright Light gt ServiceTypeLight, Brightness??!
• Brightness gt 30 (in daytime)
• Brightness gt 10 (at night)
• Alarm Clock gt ServiceTypeClock, Effect??!
• Effect Sound (in case of driving, or sleeping)
• Effect Vibration (in case of studying)
• Effect Display (in case of working)

44
Large scale semantic service discovery UbiSearch
(ICU 07)

• Proposed idea
• Semantically similar services are clustered
  around nearby resolvers in the service discovery
  overlay network
• The search space for a query is efficiently
  limited

C
D
A
B
C
G
A
B
H
D
E
F
mapping
E
service discovery resolver service
indexes semantic vector space
I
F
G
H
I
45
Overview of UbiSearch

• Features of UbiSearch
• Semantic service description and similarity
  measuring function
• Ontology based attribute-value pairs (AV-pairs)
• Semantic distance function
• Semantic association between services and
  resolvers
• Semantic vector space
• Semantic service discovery network
• Resolver organizing
• Registration query routing

46
Service Description

• Ontology based attribute-value pair (AV-pair)

ltservicegt ltidgt TC1004lt/idgt lttypegt ltnamegt
Traffic Monitoring Camera lt/namegt lt/typesgt
ltattributesgt ltnamegt location
lt/namegt ltvaluegt Broadway lt/valuegt
lt/attributegt ltattributegt
ltnamegt traffic state lt/namegt
ltvaluegt Busy lt/valuegt lt/attributegt
ltattributegt ltnamegt road condition
lt/namegt ltvaluegt Dry lt/valuegt
lt/attributegt lt/attributesgt
ltinterfacesgt lt/interfacesgt
lt/servicesgt
Examples of ontology hierarchy for locations and
traffic states. The arrows represent subsumption
relationships.
A service description for Traffic Monitoring
Camera.
47
Semantic Distance Function

• Distance function between two concepts in
  hierarchically-structured semantic network 11
• where l is the length of the shortest path
  between two concepts c1 and c2 in the semantic
  tree and h is the level of the tree of the direct
  common subsumer from c1 and c2 .
• a 0 and ß 0 are parameters scaling the
  contribution of shortest path length and depth,
  respectively.
• According to benchmark tests, the optimal
  parameters are a 0.2 and ß 0.6.
• Distance function between two services
• where n is the number of attributes, vi1 and vi2
  are the values of attribute ai of service S1 and
  S2, respectively.
• ?i is the weighting factor for attribute ai given
  by a user during query request where ? ?i is n.

48
Semantic Coordination

• Lipschitz embedding
• Defined as follow
• For a given set L l1, l2, , ln, L ? X, and
  distance function d
• the Lipschitz embedding with respect to L is the
  mapping f X ? Rn such that f(x) d(x, l1),
  d(x, l2), , d(x, ln) where L n, where n is
  dimensionality of a coordinate vector.
• To obtain the coordinate vector xi ? Rn for
  service Si, the j-th component of xi is set to
  the measured distance between service Si and
  landmark service j, for j1,,n.
• In general, the accuracy of the coordination
  improves as the number of landmarks increases.
• In our case, the distance function with ?i 1 is
  used for d.

49
Semantic Service Discovery Network

• Service Registration and Query Routing
• A service is registered with a resolver whose
  zone contains its coordinate.
• Similarly, a query is routed to the resolver