Interop97 - Gaede

Planning in Spatial Internet Marketplaces

Volker Gaede
CSIRO Mathematical and Information Sciences
GPO Box 664
Canberra, ACT 2601, Australia
Volker.Gaede@cmis.csiro.au

Notes: slides from this conference presentation can be found at http://www.wiwi.hu-berlin.de/~gaede/planning.slides.ps.gz

Recently, the idea of Internet Marketplaces has been proposed as a new form of making data and computational services available to a broad public. Like in a traditional marketplace these services are made available by providers and can be purchased by customers. In contrast to traditional marketplaces, however, customers do not receive a personal copy but pay for the use of the service. The Internet is hereby used as a communication medium for both requesting and delivering the service. This is a very important difference to on-line shopping over the Internet or electronic commerce, where the Internet is typically only used to initiate a service request and the actually delivery is by, for example, surface mail.
Since its first proposal, the idea of Internet Marketplaces has been adapted by various application domains including decision support systems, mathematical software and spatial information systems. Compared to other Internet market proposals, Spatial Internet Marketplace seem have a greater market potential since they encompass a large number of communities sharing a common interest in spatially referenced data and the associated computational services. With respect to concrete implementation of such Spatial Internet Marketplaces numerous problems have to be resolved and the SMART project currently conducted at CSIRO Mathematical and Information Sciences is a first step this direction. Within the scope of the SMART project an object-oriented design model along with a list of required actions of a Spatial Internet Marketplace has been proposed. One of the major differences of the SMART proposal in comparison with other proposals is that Spatial Internet Marketplaces should cater for some support of planning. That is, designated services accept a declarative request specification and generate an executable plan which materializes the request by combining distributed data and computational services. In other words, the system determines automatically from which site to acquire the data and where to perform certain operations. This planning procedure involves identification of suitable services, generating the corresponding requests for each of them and combining the generated results.

The motivation behind having such planning services in an Internet Marketplace is similar to database systems, where users ideally should be able submit requests without knowing any internal details and the query optimizer has to generate by means of some information an executable program. At first glance, one is tempted to think that planning in a Spatial Internet Marketplace is no different to query optimization in federated database systems, but as a closer investigations shows this is not true. In the paper, we provide a detailed comparison how planning in Internet Marketplaces differs to planning in federated database systems and discuss planning with respect to the unique requirements imposed by Spatial Internet Marketplaces. For example, potential customers in a Spatial Internet Marketplace are expected to have no or only a limited knowledge of the services available and how to access them. Instead of referring to particular services, Spatial Internet Marketplace customers simply describe what kind of services they are interested. For example, a customer should be able to express requests as the following one: Intersect the Australian crop map from 1996 with the Australian soil map. Because requests like this are expected to be the rule rather than the exception, the planning paradigm of federated database systems needs to be extended by a number of additional steps such a resource discovery step which aims at identifying providers and services (ie., resources) useful for answering a given request. In summary, Spatial Internet Marketplaces planning consists of the following steps:

Resource discovery and classification step. The planning service accepts a customer's request and by means of the information stored in the registry or made available by the services performs a first selection of qualifying services, ie., identifies classes of equivalent services.
Estimation and integration step. In a next step the planner generates for each of these equivalence classes requests and contacts the various services within each class to acquire further information (eg., cost estimates, precision). If two service classes are incompatible, the planning services tries to identify a transformation service to perform the corresponding transformation, ie., conducts another resource discovery step. Clearly, the cost induced by such a transformation service have to he added to the overall cost.
Selection step. On the basis of the returned information, the planning services rank eligible services and generates a set of alternative plans.

However, in order to generate such a plan, the planning service must have some information about the existing services and the functions provided by them. The availability and the quality of this information is the key to successful planning, which in turn requires the dynamic solution of various issues to guarantee the interoperability of the services involved in the plan execution. It is not immediately clear what kind of information is necessary for planning and how this information should be made accessible. In a constantly changing environment such as the Internet, it is to be expected that services and their interface definitions also change. In contrast to distributed spatial applications, however, providers in a Spatial Internet Marketplace are less likely to inform planning services about changes. Furthermore, once generated plans might be come invalid because of the unavailability of a certain service at run-time or the change of an interface. In such cases it is important to perform dynamic service substitution by sending the request to an equivalent service. Basically, there are two possible ways to approach the above outlined problems. In the paper we present and compare these approaches with respect to their suitability and their impact on the overall design of Spatial Internet Marketplaces.