Temporal
constraint satisfaction problems have been widely studied but only recently have
been deeply investigated problems where constraints can specify the minimum and
maximum distance between two event occurrences in terms of different time
granularities (e.g., days, weeks, business day, or
academic semester). It was assumed that multi granularity problems could be
easily converted in terms of a single granularity, which is not true, since the
problem has now been proved to be inherently more difficult.
A formalism has been introduced for
temporal constraints with granularities (TCG). For example, the constraint in
the figure, indicates that the event associated to node 2 must occur two weeks
later than the one associated to node 1, and in the same semester.
The GSTP system that we are
demonstrating at ISD’03 is a web service which provides access to the implementation
of a sound and complete algorithm for consistency of multi granularity networks. This service also provides a network solution
and a view of refined explicit and implicit constraints in terms of any of the
time granularities appearing in the network.
In
order to understand why a constraint in terms of a granularity cannot be simply
substituted by a constraint in terms of a different granularity, consider how
the constraint [1,1]day between Event1 and Event2 could
be converted in terms of hours.
The
tightest candidate constraint is [1,47]hours. Indeed, any value greater than 1 would exclude the
solution of Event1 occurring in the last hour of one day and Event2 in the
first hour of the next day; similarly, any value smaller than 47 would exclude the
solution of Event1 occurring in the first
hour of one day and Event2 in the last hour of the next day.
Thus,
can we substitute [1,1]day with [1,47]hours?
NO.
There are solutions for [1,47]hours which violate [1,1]day: if Event1 occurs at
noon of one day and Event2 is 47 hours later, the second event is 2 days after the first one!
Conversion is used in the GSTP
system only to derive implied
constraints which do not substitute the original ones, but are very useful to
refine the network as a pre-processing step for the consistency algorithm.
Specifying GSTP
networks
A remote java-based interface
enables the visual specification of a constraint network, its submission to the
constraint solver, and the analysis of the output received from the solver.
A simplified real world
example is presented during the demo: a distributed workflow process for the
delivery of products resulting from online orders.
A temporal constraint network is
designed by modeling some of the constraints involved in the process; this is
an application example where the ability to work with multi granularity constraints
can be particularly appreciated.
Constraint
Network Processing
The GSTP interface offers a
number of options to configure the request that will be sent to the GSTP web service,
and then to the constraint solver. Once the network is submitted and the
service has finished its job, the output will be available for analysis at the GSTP
interface.
The
constraint solver offers several services, the first one being a consistency
check.
The second service is TCG
refinement which proves to be particularly useful in the demo example.
The third service concerns
implicit constraints. The visualization of the results in this case may be a
problem.
The GSTP interface provides different tools: automatic
positioning of nodes, visualization of arcs without constraints, views of
selected arcs and constraints, views on specific granularities, and zooming
functions.
Finally, the demo shows the
GSTP service providing the minimum solution for the given network.