A Research Agenda for AgentBased Service-Oriented
Architectures
Prof. Michael N. Huhns
Center for Information Technology
University of South Carolina
Technology Themes for the Next 50 Years





Computation and bandwidth to burn
(for data mining and simulation)
Sensory transformation, when “things
start to think”
Lightweight infrastructure
Small world: nanotechnology + MEMs +
biosystems
Extending biology
5/31/2016 2:52:58 AM
© Michael N. Huhns
2
Highlights
Why should anyone care about Web services,
especially Semantic Web services?





Services provide higher-level abstractions for organizing
applications in large-scale, open environments, resulting
in improved productivity for developers and improved
quality for the applications
Expose business functionality (conventional Web
services)
Enable programming in the large
Enable new applications to be offered dynamically and
on-demand through composition (Semantic Web
services soon; agent-based SWS eventually)
Produce robust applications (agent-based Semantic Web
services)
5/31/2016 2:52:58 AM
© Michael N. Huhns
3
If even Dilbert knows about Web services…
…so should you!
5/31/2016 2:52:58 AM
© Michael N. Huhns
4
Requirements Imposed by a ServiceOriented Architecture
Loose coupling
 Implementation neutrality
 Flexible configurability
 Long lifetime
 Granularity
 Teams
The old abstraction for computing involved individual
computations
The new abstraction for computing involves interactions
among computations
Must capture patterns of semantic and pragmatic
constraints governing service compositions

5/31/2016 2:52:58 AM
© Michael N. Huhns
5
Great Benefits, But Will SOAs Really Take
Off?
Stage
Quantity
Representation
Use
Cost
Resultant
Apps
Now
Few, on
intranets
Syntactic (WSDL)
Manual
$0
Built quickly,
but brittle and
static
II
Many via
Internet
Keyword
semantics
Manual
composition
Fee for use,
set off-line
Robust apps
via
substitution of
services
Rich semantics of
function
Automatic
discovery &
invocation
III
IV
Active
services
5/31/2016 2:52:58 AM
Semantics of
behavior
Dynamic apps
Autonomous
negotiation
© Michael N. Huhns
Robust, selforganized
teams
6
Example of Current SOA Success
Amazon.com:
 Converted monolithic application into 100’s of
services
 Applications for customer service, selling, Amazon’s
Web pages, and hosted applications invoke the
services as needed
Enables unanticipated 3rd-party applications:
 Shopping with a camera phone (uses reviewing
service, comparable product service, and current
price service)
 An AJAX/S3 Wiki (uses the simple storage service for
code and data)
 The Great Baconizer (uses recommendation service,
i.e., “whoever bought A also bought B”) [see next page]
5/31/2016 2:52:58 AM
© Michael N. Huhns
7
“The Great Baconizer”






For example, the connection between the
book “Surely You're Joking, Mr. Feynman!"
and the DVD “Real Genius“ (11 hops):
“Surely You're Joking, Mr. Feynman!" -- R.
Feynman  “Genius: The Life and Science of
Richard Feynman," -- J. Gleick
:::
 “Weird Science“ (DVD) -- John Hughes
 “Top Secret!" (DVD) -- Val Kilmer
 “Real Genius" (DVD) -- Val Kilmer
5/31/2016 2:52:58 AM
© Michael N. Huhns
8
A Source for Research Ideas:
What Are the Limitations of the WS Triangle?
Consider each vertex and edge:
Service
Broker
Publish or
announce
(WSDL)
Service
Provider
Not well-known
5/31/2016 2:52:58 AM
Registry; well-known
Find or
discover
(UDDI)
Bind or
invoke
(SOAP)
© Michael N. Huhns
Service
Requestor
9
Needed Enhancements





For advertising and publishing, services must
be described with greater precision and
structure
A registry must be able to certify providers,
so it can endorse them
A requestor must be able to trust a registry,
which involves reputation and incentives
A registry must be able to understand the
needs of a requestor
A requestor and provider must conduct longlived flexible transactions, must establish SLA,
must monitor compliance with SLA
5/31/2016 2:52:58 AM
© Michael N. Huhns
10
Additional Limitations




A Web service knows only about itself — not
about its users, clients, or customers
Web services are not designed to use and
reconcile ontologies among each other or
with their clients
Web services are passive until invoked; they
can’t provide alerts or updates when new
information becomes available
Web services do not cooperate with each
other or self-organize, although they can be
composed by external systems
5/31/2016 2:52:58 AM
© Michael N. Huhns
11
Web Services Will Become Agents

Current Web services have a database
or a programmatic basis. Both are
unsatisfactory…
5/31/2016 2:52:58 AM
© Michael N. Huhns
12
Simple B2C Web Service Example
Suppose you want to sell cameras over the
Web, debit a credit card, and guarantee
next-day delivery
SellCamera
Web Service
 Your application must





Internet
record a sale in a sales database
Sales
Inventory
debit the credit card
Shipping
send an order to the shipping department
receive an OK from the shipping department for
next-day delivery
update an inventory database
5/31/2016 2:52:58 AM
© Michael N. Huhns
13
“Traditional” B2C Problems


What if the order is shipped, but the debit
fails?
What if the debit succeeds, but the order was
never entered or shipped?
5/31/2016 2:52:58 AM
© Michael N. Huhns
14
Database Approach
A traditional database approach works only for a closed
environment:
 Transaction processing (TP) monitors (such as IBM’s
CICS, Transarc’s Encina, BEA System’s Tuxedo) can
ensure that all or none of the steps are completed, and
that systems eventually reach a consistent state
 But what if the user’s modem is disconnected right after
he clicks on OK? Did the order succeed? What if the line
went dead before the acknowledgement arrives? Will the
user order again?
The TP monitor cannot get the user, or the user’s agent,
into a consistent state!
5/31/2016 2:52:58 AM
© Michael N. Huhns
15
Approach for Open Environment




Server application could send email about credit problems, or
detect duplicate transactions
Downloaded applet could synchronize with server after broken
connection was restored, and recover transaction; applet could
communicate using http, or directly with server objects via
CORBA/IIOP or RMI
If there are too many orders to process synchronously, they
could be put in a message queue, managed by a Message
Oriented Middleware server (which guarantees message delivery
or failure notification), and customers would be notified by
email when the transaction is complete
Server could use 2PC, IF the user’s agent understands that
protocol
The server behaves like an agent!
5/31/2016 2:52:58 AM
© Michael N. Huhns
16
Programming Basis for Web Services


Under a programming approach,
software is partitioned into
composable services, which are
invoked by an application using, for
example, RMI
Suppose application A invokes
A
service B, but B is busy and
delegates the request to service C.
When service C sends a response to
A, A fails because it expected a
response from B
5/31/2016 2:52:58 AM
© Michael N. Huhns
B
delegate
request
C
response
17
Trends…
First Generation Web
Hardware:
 Connect all PCs
Software:
 Create a Web page
for everybody and
everything
Use Google to get information, e.g., population
of China
How do we get correct information?
There is a power in consensus that is not being
exploited!
5/31/2016 2:52:58 AM
© Michael N. Huhns
18
Correct Grammar from Google


Is the correct phrase “in the West
Coast” or “on the West Coast” or “at the
West Coast”?
In Google, the first yields 0.6M hits, the
second 10.2M hits, and the third 0.2M
hits
5/31/2016 2:52:58 AM
© Michael N. Huhns
19
Trends…
Second Generation Web
Hardware:
 Connect all devices,
sensors, objects,
(and PCs)
Software:
 Create a Web
service for each
Use Semantic Web (OWL-S) to glue
behaviors into a composite; this is
programming-in-the-large (or SOC)
How do we get our desired aggregate behavior?
5/31/2016 2:52:58 AM
© Michael N. Huhns
20
Engineering a SOC System
Goal
Subgoal
AI planning leads
to a declarative
description of
what is desired…
Subgoal
Subgoal
Subgoal
Subgoal
Subgoal
Subgoal
Subgoal
Subgoal
Discovering, Matching, Planning, Composing
Repository/Directory of Services
WebService1
AI planning leads
to a procedural
description of
how to achieve
what is desired…
WebService2
...
WebServiceN
Discovering, Matching, Planning, Composing
Task4
Task2
Task6
Task5
Task1
Task7
Task3
5/31/2016 2:52:58 AM
© Michael N. Huhns
21
Composed Web Services
Currency
Converter
Web Service
Stock Quote
Web Service
1. WSDL
“IBM”
3. WSDL
“$100”
2. “$100”
4. “¥800”
Client
Application
5/31/2016 2:52:58 AM
© Michael N. Huhns
22
WSDL-P: Next-Generation Composition
Stock Quote
Web Service
2. WSDL-P
“$100” +
OWL-S
Currency
Converter
Web Service
1. WSDL-P
“IBM”+
OWL-S
3. “¥800”
Client
Application
An OWL-S (or BPEL4WS) description of the
workflow is communicated through the services
5/31/2016 2:52:58 AM
© Michael N. Huhns
23
Complementary Features
Service-Oriented
Multiagent-Based
Computing
Computing
 Encapsulation
 Encapsulation
 Autonomy
 Autonomy
 Distribution
 Distribution
 Interaction via
 Interaction via
messaging
messaging
 Mutual understanding  Proactivity
 Ontologies for objects
 Negotiation ability
and processes
5/31/2016 2:52:58 AM
© Michael N. Huhns
24
Service Interaction Process
5/31/2016 2:52:58 AM
© Michael N. Huhns
25
Use Cases for Engagement

Simple query-reply protocol:

Simple request-reply protocol:
Requestor: "Please provide your service for me.“
Requestor: "Will you provide your service for me?“
Provider: "OK.“
Requestor: "Will you provide your service for me?“
Provider: "No.“

Negotiate commitment protocol:
Requestor: "Will you provide your service for me?“
Provider: "Only if you pay me ten dollars.“
Requestor: "I will pay you five dollars if you provide your service
by 5:00 pm.“
Provider: "OK.“
Results: an explicit contract for the service, and commitments
by the parties to satisfy the terms of the contract.
5/31/2016 2:52:58 AM
© Michael N. Huhns
26
Engagement: Commitment Protocol
5/31/2016 2:52:58 AM
© Michael N. Huhns
27
Trends…
Third Generation Web
Hardware:
 Now that
everything is
connected…
Software:
 Create an agent to wrap
and represent every Web
page and Web service
Use ??? to create new behaviors dynamically and
form a new consensus about each; e.g., who
programs the lights at a highway intersection?
How do we get consistent and correct behavior?
5/31/2016 2:52:58 AM
© Michael N. Huhns
28
Coherence and Commitments


Coherence is how well a system
behaves as a unit. It requires some
form of organization, typically
hierarchical
Social commitments among agents
are a means to achieve coherence. An
agent’s commitment to another agent
or to its society




Is unidirectional
Arises within a well-defined scope or
context
Is revocable with restrictions
Enables coordination through the ordering
and occurrence of actions by the agents
5/31/2016 2:52:58 AM
© Michael N. Huhns
29
Commitments for Contracts
Commitments capture contracts. Importantly,
commitments are
 Public (unlike beliefs and intentions)
 Can be used as the basis for compliance
 Contracts apply between parties, in a context
 Other approaches:



Are single-agent focused, e.g., deontic logic
Don’t handle organizational aspects of contracts
Don’t accommodate manipulation of contracts
5/31/2016 2:52:58 AM
© Michael N. Huhns
30
Manipulating Commitments

Operations on commitments:







Create
Discharge (satisfy)
Cancel
Release (eliminate)
Delegate (change debtor)
Assign (change creditor)
Metacommitments constrain the
manipulation of commitments
5/31/2016 2:52:58 AM
© Michael N. Huhns
31
Commitment Protocols


Protocols enable open systems to be
constructed
Interaction protocols expressed in terms of




Participants’ commitments
Actions for performing operations on commitments
(to create and manipulate them)
Constraints on the above, e.g., captured in
temporal logic
Examples: escrow, payment, RosettaNet (107
request-response PIPs)
5/31/2016 2:52:58 AM
© Michael N. Huhns
32
Compliance with Protocols
In an open environment, agents are
contributed by different vendors and
serve different interests
 How can an application check if the
agents comply with specified
protocols?



Coordination aspects: traditional techniques
Commitment aspects: representations of
the agents’ commitments in temporal logic
Commitment protocols are specified in
terms of



Main roles and sphere of commitment
Roles essential for coordination
Domain-specific propositions and actions
5/31/2016 2:52:58 AM
© Michael N. Huhns
33
Verifying Compliance with Commitment
Protocols

Specification



models based on potential causality
commitments based on branching-time TL
Run-time Verification



respects design autonomy
uses TL model-checking
local verification based on observed
messages
5/31/2016 2:52:58 AM
© Michael N. Huhns
34
The Power of Consensus…
For Accurate Services
Numerous Web services are being deployed and
many of them have similar, but not identical
behavior. For example, you might have a
software application that needs accurate and
timely stock quotations, but one stock service
might provide more frequent but less
accurate results than another.
How do we get consensus behavior?
5/31/2016 2:52:58 AM
© Michael N. Huhns
35
Long-Term Research Objectives


Robust software constructed out of large
numbers of agents with redundant
functionality
Consensus software for societal systems



Anyone can publish information as Web pages
Anyone should be able to contribute to the
behavior of societal systems, such as for power
distribution, traffic control, telecommunications
regulation, etc.
Large-scale ethical and philosophical agent
societies
5/31/2016 2:52:58 AM
© Michael N. Huhns
36
Robustness through Multiple Viewpoints




Errors will always be in complex systems
Errors can be detected and corrected through
redundancy
Hypothesis: agents are the right level of
granularity and abstraction for adding
redundancy
Agents can vote, present different viewpoints,
cover for agents that misbehave, provide
checks and balances
5/31/2016 2:52:58 AM
© Michael N. Huhns
37
Redundancy Is the Basis for Robustness
CREDIT
REPORT
AGENCY
CREDIT
REPORT
AGENCY
Receive
application
Collect
applicant’s info
Verify
applicant’s info
CREDIT
REPORT
AGENCY
Obtain applicant’s
credit score
[Credit score>600] Approve
credit
[Credit score<=600]
Deny credit
5/31/2016 2:52:58 AM
© Michael N. Huhns
38
Advanced Composition

Suppose an application needs simply to sort
some data items, and suppose there are 5
Web sites that offer sorting services described
by their input data types, output date type,
time complexity, space complexity, and
quality:
Sort1





Application
One is faster
?
Sort2
One handles more data types
?
? ?
One is often busy
Sort3
Sort5
Sort4
One returns a stream of results,
another a batch
One costs less
5/31/2016 2:52:58 AM
© Michael N. Huhns
39
Advanced Composition

Possible approaches







(cont.)
Application invokes services randomly until one succeeds
Application ranks services and invokes them in order until
one succeeds
Application invokes all services and reconciles the results
Application contracts with one service after requesting bids
Services self-organize into a team of sorting services and
route requests to the best one
The last two require that the services behave like
agents
The last two are scalable and robust
5/31/2016 2:52:58 AM
© Michael N. Huhns
40
Agent-Based N-Version Robustness
Each agent would have to know its role in this
system as well as:






Something about its own algorithm, such as its
time/space complexity, and input/output data
structures
Something about other agents, such as their time
and space complexity and reliability
How to negotiate
How to communicate
How to compare results
How to manage reputations and trust
5/31/2016 2:52:58 AM
© Michael N. Huhns
41
Architecture for Agent-Based N-Version Robustness
Single Task
Choose Algorithm based on: (1) data type, (2) time & space constraints
Algorithm #1
Algorithm #2
Algorithm #N
Compare Results and Select Best
Single Result
5/31/2016 2:52:58 AM
© Michael N. Huhns
42
Experimental Results



We asked 25 students to each write
sorting and list-reversing algorithms
Each algorithm was converted to an
agent, with a wrapper written in Jade
The results show that


5/31/2016 2:52:58 AM
The same wrapper can be used for
algorithms in different domains
More agents (algorithms) produce better
results than any one alone
© Michael N. Huhns
43
Summary
Multiagent System Properties
Benefits for System Development
Autonomous, objective-oriented
behavior; agent-oriented decomposition
Autonomous, active functionality that
adapts to the users’ needs; reuse of whole
subsystems and flexible interactions
Dynamic composition and
customization
Scalability
Interaction abstractions; statistical or
probabilistic protocols
Friction-free software; open systems;
interactions among heterogeneous
systems; move from sophisticated and
learned e-commerce protocols to dynamic
selection of protocols
Multiple viewpoints, negotiation, and
collaboration
Robustness and reliability
Social abstractions
High-level modeling abstractions
5/31/2016 2:52:58 AM
© Michael N. Huhns
44
Advertisement for a Textbook…
Munindar Singh and Michael Huhns:
“Service-Oriented Computing:
Semantics, Processes, Agents,”
John Wiley & Sons, Inc., 2005.
5/31/2016 2:52:58 AM
© Michael N. Huhns
45