Automated QoS-Aware Service Selection Based on Soft Constraints

KEIS, ELIAS; LOPEZ POMBO, CARLOS GUSTAVO; MARTINEZ SUÑÉ, AGUSTÍN ELOY; KNAP, ALEXANDER

Aveiro

Workshop; Workshop on Algebraic Development Techniques - WADT 2022; 2022

University of Aveiro

In software-as-a-service paradigms such as Service-oriented Computing – SOC, software systems are no longer monolithic chunks of code executing within the boundaries of an organisation. On the contrary, this new generation of applications run over globally available computational resources and rely on a dedicated middleware responsible for the discovery of services. Such services are bound at runtime, subject to the negotiation of a Service Level Agreement– SLA, so they can collectively fulfill a given business goal [2]. Providers of cloud computing platforms rely on these concepts and offer a high degree of customisation for their services, hence companies can configure the resources to better suit their business needs. Well-known examples of such customisation are pricing schemes that depend on the amount of time a computational resource is used. An essential aspect of such a service selection (known as the Service Selection Problem [1, pt. II]) is determining whether the Quality of Service (QoS) profile of a service (i.e., a description of the potential values that its quantitative attributes might adopt) satisfies the QoS requirements of a client. The interested reader is pointed to [6] for a comparative review of existing approaches and [4] for a systematic literature review of the problem. We are interested in this problem in the context of composite services that can be described as workflows. As shown in the next diagram, a workflow consists of one or multiple tasks (squared nodes) composed in different ways. Each task induces a class of concrete services (rounded nodes) that can fulfill its functional needs. Composition can be sequential, parallel, by a choice, or by a bounded loop. One possible approach to solving this problem can be derived from our solution to the service selection problem for single service requirements [5] by indepedently selecting the best service candidate for each task given local requirements for such task. If the use of workflows comes with an interest in global QoS requirements that must be satisfied by the chosen services for all the tasks in the workflow, then choosing the best candidate for each task might not satisfy the global goals. It is known that adding hard constraints reduces the number of matching solutions and might naturally lead to not having any solution left due to an overconstraint problem. We aim at overcoming this drawback by primarily focusing on the satisfaction of the global requirements for the aggregated values of the QoS attributes. Our proposal uses Constraint Programming (CP) [7] to automate the process of selecting adequate services based on their QoS properties, and the use of soft constraints comes in handy as the solver can omit them if the constraint satisfaction problem (CSP) [3] is unsatisfiable because it is overconstrained.In this work, we present a tool for solving the service selection problem for composite services in a soft way. We leverage on MiniBrass, a tool presented by Schiendorfer et al. in [8] that extends the MiniZinc constraint modeling language and tool, providing various options to model and solve soft CSPs based on the unifying algebraic theory of partial valuation structures. Specifically, our tool provides the means for: 1) describing a service workflow over which the service selection has to be performed, 2) expressing QoS profiles associated with concrete services as constraints over its QoS attributes, 3) expressing QoS requirements as soft constraints over the aggregated value of QoS attributes over the execution of the workflow, 4) automatically finding the best (if any) assignment of services to tasks given the above set up.