Decomposing the Service Composition Problem

Abstract

Many approaches to the Web service composition problem benefit from their use of formal methods to guarantee the correctness of the composite services that they produce, but these approaches often require the functionality of the composite service to be specified using one particular formalism (e.g., goal graphs, temporal logic, pre-/post-conditions). As a result, each of these existing approaches falls short in realizing a composite service when the required functionality cannot be fully expressed in the supported formalism. To overcome this problem, we propose a new formal meta-framework that (a) is capable of reusing any existing formalisms and (b) allows the use of functional requirements that are currently not expressible in any one formalism. Our technique assumes that any functional requirement can be decomposed and expressed as a boolean combination of "atomic" requirements, which are representable in at least one existing formalism. Based on this assumption, we use existing methods to identify sets of Web services that conform to the atomic requirements. Our meta-framework then identifies compositions that conform to the overall functional requirement by (a) employing satisfiability techniques to prune the (exponentially large) space of possible compositions and (b) building workable compositions from satisfiable sets of services. As a result, our meta-framework allows for easy and effective memorization of prior composition results, thereby enhancing the efficiency of generating new compositions.

Reasoning with qualitative preferences for optimization of component-based system development

“…However, each of these approaches to the service composition problem specifies one type of semantics that must be used to interpret the descriptions of the available component services and the desired functionality of the composition. In [59], we introduced a new "metaframework" for service composition that supports the use of multiple formal semantics, i.e., more than one specification language, to specify and verify functional requirements for a composite service. This meta-framework, which forms part of the theoretical foundation of the component-based system development framework described in this dissertation, is presented in greater detail in Chapter 4.…”

“…In [53], the "type-aware" Web service composition problem, where the functionality specified by a goal service must be provided and a set of type-correctness constraints must be satisfied, is reduced by Nam et al to an instance of the Boolean satisfiability problem (commonly known as the SAT problem for short) by taking advantage of a type hierarchy that is included in the problem definition. To the best of our knowledge, this is the only existing approach other than our "meta-framework" in [59] that directly reduces a type of Web service composition to Boolean satisfiability. Unfortunately, the framework in [53] considers only properties that can be expressed as inputs or outputs of the goal service, and it forms compositions based only on compatibility between services.…”

“…However, this problem is known to be NP-hard for non-functional properties with quantitative valuations [3]. Composition methods that consider non-functional properties also tend to lack the flexibility that our meta-framework from [59] provides in allowing the use of different formalisms to specify requirements and appropriate composition techniques.…”

“…In contrast, Santhanam et al [74] use qualitative valuations to select services to compose based on a set of preferences among NFPs expressed by the stakeholders in a different language for qualitative preferences, namely tradeoff-enhanced conditional preference networks (TCPnets) [14]. This approach informed our extension of our previous "meta-framework" [59] to handle global qualitative preferences over non-functional properties [62]. After decomposing the functional requirements for a composition as in [59], our method in [62] then applies NFP preferences to perform local selection of preferred component services.…”

“…This approach informed our extension of our previous "meta-framework" [59] to handle global qualitative preferences over non-functional properties [62]. After decomposing the functional requirements for a composition as in [59], our method in [62] then applies NFP preferences to perform local selection of preferred component services. The result is a service composition framework that will always identify a composition which satisfies functional requirements and is at least as preferred as any other composition that satisfies the same functional requirements, as long as all component services being considered are compatible with each other.…”