Model-Based Extraction of Knowledge about the Effect of Cloud Application Context on Application Service Cost and Quality of Service

Abstract

With the increased usage of cloud computing in production environments, both for scientific workflows and industrial applications, the focus of application providers shifts towards service cost optimisation. One of the ways to achieve minimised service execution cost is to optimise the placement of the service in the resource pool of the cloud data centres. An increasing number of research approaches is focusing on using machine learning algorithms to deal with dynamic cloud workloads by allocating resources to services in an adaptive way. Many of such solutions are intended for cloud infrastructure providers and deal only with specific types of cloud services. In this paper, we present a model-based approach aimed at the providers of applications hosted in the cloud, which is applicable in early phases of the service lifecycle and can be used for any cloud application service. Using several machine learning methods, we create models to predict cloud service cost and response times of two cloud applications. We also explore how to extract knowledge about the effect that the cloud application context has on both service cost and quality of service so that the gained knowledge can be used in the service placement decision process. The experimental results demonstrate the ability of providing relevant information about the impact of cloud application context parameters on service cost and quality of service. The results also indicate the relevance of our approach for applications in preproduction phase since application providers can gain useful insights regarding service placement decision without acquiring extensive training datasets.

Model-Based Cloud Service Deployment Optimisation Method For Minimisation of Application Service Operational Cost

“…The metamodel parameters were selected so that they can be applied on any application service deployable in a cloud environment, and as such enable the generalisation of the proposed optimisation approach required by R3. The details of the metamodel development and the chosen parameters are described in [56] and Section 3.3. The last requirement, R4, was tackled by the deployment optimisation algorithm which provides optimised service placement decision based on the simulation of application service execution in given cloud deployment environments, using the model of cloud application and its context while taking into account the QoS requirements defined by the service level agreement.…”

“…Data is obtained either by performing measurements for an already implemented service to which the method is being applied or, in case there is no possibility to perform measurements, by using existing data of a similar service that can be used for an estimation of the optimised deployment option for a service that is not yet implemented. The measurement process applied to the chosen application service use cases is provided in detail in [56], where it is described together with the data analysis, which will also be briefly presented in Section 4 of this paper. The obtained data includes metrics and parameters based on the cloud application context metamodel, defined in the next section.…”

“…From the perspective of a cloud application service provider, crucial parts of the management of a cloud application in its operative state are the cost of running the application service in the cloud and making sure that it delivers the appropriate level of service quality. There are several aspects of the cloud application service and its deployment environment that affect the QoS and cost of application execution in a cloud environment; we refer to those as the cloud application context [56]. This section introduces the cloud application context metamodel and explains the selection of parameters used in the proposed metamodel.…”

“…User types are based on the actions that users perform and generate certain load which affects the application's utilisation of resources. Further details on application user types, examples, and descriptions of user types analyzed in validation use cases for optimisation method presented in this paper can be found in [56]. Another application-related parameter is the Service Level Agreement (SLA), which defines the QoS requirements, commonly through a set of Service Level Objectives (SLOs), that will guarantee a level of service quality delivered to the application end users.…”

“…Permutation importance was introduced by Breiman in [83] where the method is described for random forest models, but the technique applies to regression as well. In validation of optimisation method proposed in this paper, we use permutation importance analysis on the regularised regression models [56]. Permutation importance is calculated after the model has been fitted.…”