Conversation-based specification
and composition of agent services
Bao Quoc Vo
(joint work with Lin Padgham)
Intelligent Agent Groups
RMIT University
Conversation
• A Web service conversation involves
maintaining some state during an
interaction that involves multiple
messages and/or participants.
• From Web Services Glossary (2004-02-11)
Conversations and protocols
1: requestTransaction
2: enterPIN
Customer
3: authorised
4: requestTransfer
ATM
Protocol models
• Perspective
– Global (cooperative) vs. service-centric
– The latter usually requires less constraints on
protocol compatibility
• Formalism
– Sequence diagrams
– FSMs/statecharts
– Process-based models
• Expressive power
Conditions, State management, Correlation,
Transactions, Modularization, Exception handling,
etc.
Our model of WS/agent services
• A service is an event-driven component
– events are sending and receiving of messages
• service description (external schema)
comprises
– external interface
– system dynamics (interaction protocol):
<S> : <dir><msg>  <S>
• <S> is the symbolic name of a state;
• <dir> is the direction of the message (?/!)
• <msg> is the message name
Example
(an online banking system)
Service Banking {
Interface {
RECEIVE enterPIN(Account acc, EncryptedPIN PIN);
RECEIVE requestTransfer(Account toAcc, float amount);
RECEIVE requestBalance();
SEND invalidPIN();
SEND authorised();
SEND overdrawn();
SEND transactionApproved();
SEND currentBalance(float balance); };
Protocol {
States { 0(init,final), 1, 2, 3, 4, 5(final), 6(final) };
Transitions {
0 : ?enterPIN
-> 1;
1 : !invalidPIN
-> 0;
1 : !authorised
-> 2;
2 : ?requestTransfer -> 4;
2 : ?requestBalance
-> 3;
3 : !currentBalance
-> 0;
4 : !transferApproved -> 5;
4 : !overdrawn
-> 6; };
};
};
Example (2)
Service Composition
• A composition system consists of the external
schema of the target service plus the external
schemata of the available component services,
C = (ST, {S1, …, Sn})
• The composition problem:
constructing the internal model of the target service
so that it employs only messages of the component
services and satisfies the interaction protocols.
• Well-defined compositions:
free of deadlocks and unspecified receptions.
Service Composition (2)
• A realisation of a composite service ST within a
composition system (ST, {S1, …, Sn}) is a FSM
M = (Q,Σ, δ, q0, F)
• Informally, M simulates the behaviour of ST as
follows:
– The states in ST are mapped to a subset of states of M.
– Each receive message m in ST is simulated in M
originated from the corresponding state(s) of M.
– Each send message m in ST corresponds to a subgraph
G of M that implements the transactions needed to
produce the data carried by m. The final transactions in
G are also labelled m pointing to the corresponding
state(s) of M.
Service Composition (3)
• The following will be interacting with M: the components
services S1, …, Sn and a well-defined client which is a
complement of ST (denoted C(ST)).
• During execution of the composite service, the whole
machinery M = (M,<S1, …, Sn ,C(ST)>) moves together
through the composition states, producing execution
traces; all execution traces comprise execution tree T.
• Unspecified reception: A node on T where either M
sends out a message m and none of S1, …, Sn ,C(ST) is in
a state to receive m, or one of S1, …, Sn ,C(ST) is in a
state of not receiving any message and only sending out
a message m and M is not in a state to receive m.
• Free of deadlock: All finite branches on T end at a
composition state containing only final states of
M, S1, …, Sn ,C(ST).
Service Composition (4)
• A realisation M = (Q,Σ, δ, q0, F) of a
composition system C is an internal
model of C if M is free of unspecified
receptions and free of deadlocks.
• Theorem:
Checking whether a realisation M is an
internal model of a given composition
system C is poly-time.
Asynchronous communication
• Under a synchronous semantics, the finitestate machines describing the protocols of
the component services and the internal
model of the composite service are required
to advance atomically.
• To allow asynchronous communication, FIFO
channels containing messages exchanged
between different components of the system
are introduced.
Asynchronous communication (2)
• A machinery M now contains also the
communication medium state of the
systems (i.e. the content of the FIFO
channels)
• A configuration consists of the
composition state and the medium
state.
• Deadlock and unspecified receptions
are accordingly defined.
Synthesising composition
• Ontological reasoning: mapping
between component interfaces
• Example:
–
–
–
–
–
ATS :: requestFlight.flightID  AL :: flightRequest.flightID
AL :: pricingInfo.price  ATS :: offer.price
forward AL :: flightUnavailable as ATS :: notAvailable
forward ATS :: ack as AL :: bookFlight
ATS :: requestFlight.flightID  AL :: bookFlight.flightID
Synthesising composition (2)
• Constructing execution tree: exploring
all possible execution traces given the
finite state machines of the component
services and the target service
• error nodes on the execution trees:
corresponding to states in which a
deadlock or unspecified reception might
arise. (Theorem 2 in the paper.)
Conclusion
• We tackling the problem of service
composition in both the synchronous and
asynchronous communication cases.
• Given the specifications of the component
services and the target service, we construct
the internal model for the target service
satisfying desirable properties.
• Failure handling and allowing services to
effect the physical world is the future work.