Alessio Gambi

Negotiation of Service Level Agreements: An Architecture and a Search-Based Approach

Software systems built by composing existing services are more and more capturing the interest of researchers and practitioners. The envisaged long term scenario is that services, offered by some competing providers, are chosen by some consumers and used for their own purpose, possibly, in conjunction with other services. In the case the consumer is not anymore satisfied by the performance of some service, he can try to replace it with some other service. This implies the creation of a global market of services and poses new requirements concerning validation of exploited services, security of transactions engaged with services, trustworthiness, creation and negotiation of Service Level Agreements with these services. In this paper we focus on the last aspect and present our approach for negotiation of Service Level Agreements. Our architecture supports the actuation of various negotiation processes and offers a search-based algorithm to assist the negotiating parts in the achievement of an agreement.

Towards the formalization of properties of cloud-based elastic systems

Cloud-based elastic systems run on a cloud infrastructure and have the capability of dynamically adjusting the allocation of their resources in response to changes in the workload, in a way that balances the trade-off between the desired quality-of-service and the operational costs. The actual elastic behavior of these systems is determined by a combination of factors, including the input workload, the logic of the elastic controller determining the type of resource adjustment, and the underlying technological platform implementing the cloud infrastructure. All these factors have to be taken into account to express the desired elastic behavior of a system, as well as to verify whether the system manifests or not such a behavior. In this paper, we take a first step into these directions, by proposing a formalization, based on the CLTL^t(D) temporal logic, of several concepts and properties related to the behavior of cloud-based elastic systems. We also report on our preliminary evaluation of the feasibility to check the (formalized) properties on execution traces using an automated verification tool.

Iterative test suites refinement for elastic computing systems

Elastic computing systems can dynamically scale to continuously and cost-effectively provide their required Quality of Service in face of time-varying workloads, and they are usually implemented in the cloud. Despite their wide-spread adoption by industry, a formal definition of elasticity and suitable procedures for its assessment and verification are still missing. Both academia and industry are trying to adapt established testing procedures for functional and non-functional properties, with limited effectiveness with respect to elasticity. In this paper we propose a new methodology to automatically generate test-suites for testing the elastic properties of systems. Elasticity, plasticity, and oscillations are first formalized through a convenient behavioral abstraction of the elastic system and then used to drive an iterative test suite refinement process. The outcomes of our approach are a test suite tailored to the violation of elasticity properties and a human-readable abstraction of the system behavior to further support diagnosis and fix.

Modeling cloud performance with kriging

Cloud infrastructures allow service providers to implement elastic applications. These can be scaled at runtime to dynamically adjust their resources allocation to maintain consistent quality of service in response to changing working conditions, like flash crowds or periodic peaks. Providers need models to predict the system performances of different resource allocations to fully exploit dynamic application scaling. Traditional performance models such as linear models and queueing networks might be simplistic for real Cloud applications; moreover, they are not robust to change. We propose a performance modeling approach that is practical for highly variable elastic applications in the Cloud and automatically adapts to changing working conditions. We show the effectiveness of the proposed approach for the synthesis of a self-adaptive controller.

Assurance of Self-adaptive Controllers for the Cloud

In this paper we discuss the assurance of self-adaptive controllers for the Cloud, and we propose a taxonomy of controllers based on the supported assurance level. Self-adaptive systems for the Cloud are commonly built by means of controllers that aim to guarantee the required quality of service while containing costs, through a careful allocation of resources. Controllers determine the allocation of resources at runtime, based on the inputs and the status of the system, and referring to some knowledge, usually represented as adaptation rules or models. Assuring the reliability of self-adaptive controllers account to assuring that the adaptation rules or models represent well the system evolution. In this paper, we identify different categories of control models based on the assurance approaches. We introduce two main dimensions that characterize control models. The dimensions refer to the flexibility and scope of the system adaptability, and to the accuracy of the assurance results. We group control models in three main classes that depend on the kind of supported assurance that may be checked either at design or runtime. Controllers that support assurance of the control models at design time privilege reliability over adaptability. They usually represent the system at a high granularity level and come with high costs. Controllers that support assurance of the control models at runtime privilege adaptability over reliability. They represent the system at a finer granularity level and come with reduced costs. Controllers that combine different models may balance verification at design and runtime and find a good trade off between reliability, adaptability, granularity and costs.

Engineering autonomic controllers for virtualized web applications

Modern Web applications are often hosted in a virtualized cloud computing infrastructure, and can dynamically scale in response to unpredictable changes in the workload to guarantee a given service level agreement. In this paper we propose to use Kriging surrogate models to approximate the performance profile of virtualized, multi-tier Web applications. The model is first built through a set of automated and controlled experiments at staging time, and can be later updated and refined by monitoring the Web application deployed in production. We claim that surrogate modeling makes a very good candidate for a modeldriven approach to the engineering of an autonomic controller. Our experimental evaluation shows that the model predictions are faithful to the observed systems performance, they improve with an increasing amount of samples and they can be computed quickly. We also provide evidence that the model can be effectively used to synthetize an aggregated objective function, a critical component of the autonomic controller. The approach is evaluated in the context of a RESTful Web service composition case study deployed on the RESERVOIR cloud.

SLA Protection models for virtualized data centers

Enterprise services must satisfy strong requirements that are coded in agreements with customers, commonly called service level agreements (SLA). To satisfy SLAs in critical conditions, conventional data centers are often greatly over-dimensioned, wasting resources and raising service costs. Virtualized data centers (VDC) provide an opportunity to significantly reduce over-dimensioning, and so reduce service costs without negatively affecting service agreements, through dynamic adaptation. In this paper, we discuss the problems involved in creating self-adaptive enterprise services in virtualized data centers, and we investigate solution strategies. We envision a set of models that help adaptation controllers to identify suitable reactions to changes in service level agreement and environmental execution conditions. We introduce models at different abstraction levels, to support the evaluation of the impacts of adaptation actions on system and SLA. We explore the requirements and specify the characteristics of these models through a case study: a Video on Demand service delivered using VDCs.

CoMoT -- A Platform-as-a-Service for Elasticity in the Cloud

Platform-as-a-Service (PaaS) should support the design, deployment, execution, test and monitoring of native elastic systems constructed from elastic service units based on multi-dimensional elasticity requirements. In this paper, we discuss fundamental building blocks for enabling multi-dimensional elasticity programming of software-defined elastic systems. We describe CoMoT, a novel PaaS for elasticity in the cloud that is developed based on these fundamental building blocks.

Kriging-Based Self-Adaptive Cloud Controllers

Cloud technology is rapidly substituting classic computing solutions, and challenges the community with new problems. In this paper we focus on controllers for cloud application elasticity, and propose a novel solution for self-adaptive cloud controllers based on Kriging models. Cloud controllers are application specific schedulers that allocate resources to applications running in the cloud, aiming to meet the quality of service requirements while optimizing the execution costs. General-purpose cloud resource schedulers provide sub-optimal solutions to the problem with respect to application-specific solutions that we call cloud controllers. In this paper we discuss a general way to design self-adaptive cloud controllers based on Kriging models. We present Kriging models, and show how they can be used for building efficient controllers thanks to their unique characteristics. We report experimental data that confirm the suitability of Kriging models to support efficient cloud control and open the way to the development of a new generation of cloud controllers.

On estimating actuation delays in elastic computing systems

Elastic controllers autonomically adjust the allocation of resources in cloud computing systems. Usually such controllers assume that control actions will take immediate effect. In clouds, however, actuation times may be long, and the controllers can hardly guarantee acceptable levels of service if they neglect these actuation delays. Therefore, the ability to correctly estimate the time that control actions take effect on the systems is crucial. However, detecting actuation delays in elastic computing systems is challenging because cloud systems provide only inaccurate and incomplete data about reconfigurations timing. In this paper, we tackle the problem of estimating the delay of control actions in elastic systems. We identify recurring types of changes in the monitored metrics and requirements to properly carry out the estimation. Based on that, we develop a novel framework for the actuation delays estimation that utilizes change point detection techniques. We conduct several experiments with real-world systems to illustrate the feasibility and applicability of our framework.