Cloud computing offers the potential to dramatically reduce the cost of software services through the commoditization of information technology assets and on-demand usage patterns. However, the complexity of determining quality of service (QoS) requirements for applications in such environments introduces significant market inefficiencies and has driven the emergence of a new class of service engineering tools within the Platform-as-a-Service (PaaS) layer for modelling, analysing and planning the QoS of service based applications deployed within the cloud.
Today, Infrastructure-as-a-Service (Iaas) QoS offerings are expressed in low level terms (i.e. machinelevel, CPU speed, disk space, etc). Their customers, typically application users, are often interested in application-level parameters because the application is the thing that gives the customer the value (e.g. CFD simulation or video rendering). Therefore, the gap between the terms the Infrastructure provider offers and what the users really want is large which results in a complex relationship between application performance and resource parameters. The complexity of this relationship is increased for applications deployed across federated clouds where even low-level resource descriptions may differ due to lack of standardisation.
Service engineering techniques aim to provide IaaS customers with a set of generic tools that can manage the complexity of the relationship. However, the parameter space used to determine an optimal set of resources requested by a customer in Service Level Agreements (SLAs) is too large to estimate results with acceptable levels of accuracy and precision unless very specific models are developed for each application. This course is not economically viable for most applications. We hypothesise that if IaaS providers raise the level of abstraction for resource QoS terms used in SLAs based on benchmark scores for specific classes of applications, significant overall efficiencies will be achieved for all cloud stakeholders due increased accuracy of requirements achieve by a simplification of service planning and adaptation models, and increased market adoption and flexibility due to the simplification of the federation between Platform and Infrastructure stakeholders. The objective of this experiment is to investigate our hypothesis through the deployment of a serviceoriented application on a cloud testbed that incorporates novel PaaS engineering tools from a leading Internet of Services (IoS) project (EU IST IRMOS) and IaaS provided by BonFIRE but configured with service offers at the identified level of abstraction.
The research questions the experiment will address include: Does the expression of IaaS parameters in terms of application class benchmarks simplify the creation of application-level QoS that can be easily understood by users? How does specification of QoS in application-level terms provide efficiencies for users and providers in a service marketplace? The experiment will not only address the specific research challenge detailed above, but will also provide a concrete exemplar scenario where research results from an IoS project can exploit FIRE, can go beyond what is possible within the current project and provide driving requirements for the FIRE facility.