Run-time Models for Self-managing Systems and Applications

Motivated by emerging trends and applications such as autonomic computing, this chapter presents an overview of some research in the stochastic analysis and optimization of multiserver systems. Our primary focus is on multiserver systems in general, since this research provides the mathematical methods and results that have been and will continue to be used for the stochastic analysis and/or optimization of existing and future multiserver systems arising in a wide variety of application domains including autonomic computing.

Applications and services delivered through large Internet Data Centers are now feasible thanks to network and server improvement, but also to virtualization, dynamic allocation of resources and dynamic migrations. The large number of servers and resources involved in these systems requires autonomic management strategies because no amount of human administrators would be capable of cloning and migrating virtual machines in time, as well as re-distributing or re-mapping the underlying hardware. At the basis of most autonomic management decisions, there is the need of evaluating own global behavior and change it when the evaluation indicates that they are not accomplishing what they were intended to do or some relevant anomalies are occurring. Decisions algorithms have to satisfy different time scales constraints. In this chapter we are interested to short-term contexts where runtime prediction models work on the basis of time series coming from samples of monitored system resources, such as disk, CPU and network utilization. In similar environments, we have to address two main issues. First, original time series are affected by limited predictability because measurements are characterized by noises due to system instability, variable offered load, heavy-tailed distributions, hardware and software interactions. Moreover, there is no existing criteria that can help us to choose a suitable prediction model and related parameters with the purpose of guaranteeing an adequate prediction quality. In this chapter, we evaluate the impact that different choices on prediction models have on different time series, and we suggest how to treat input data and whether it is convenient to choose the parameters of a prediction model in a static or dynamic way. Our conclusions are supported by a large set of analyses on realistic and synthetic data traces.

The complexity of Information Technology (IT) systems is steadily increasing and system complexity has been recognised as the main obstacle to further advancements of IT. This fact has recently raised energy management issues. Control techniques have been proposed and successfully applied to design Autonomic Computing systems, trading-off system performance with energy saving goals. As users behaviour is highly time varying and workload conditions can change substantially within the same business day, the Linear Parametrically Varying (LPV) framework is particularly promising for modeling such systems. In this chapter, a control-theoretic method to investigate the trade-off between Quality of Service (QoS) requirements and energy saving objectives in the case of admission control in Web service systems is proposed, considering as control variables the server CPU frequency and the admission probability. To quantitatively evaluate the trade-off, a dynamic model of the admission control dynamics is estimated via LPV identification techniques. Based on this model, an optimisation problem within the Model Predictive Control (MPC) framework is setup, by means of which it is possible to investigate the optimal trade-off policy to manage QoS and energy saving objectives at design time and taking into explicit account the system dynamics.

Complex pervasive systems are typically composed of a large number of heterogeneous nodes, pervasively distributed across the environment. These systems pose several new challenges such as the need for nodes to autonomously and dynamically manage themselves in order to achieve some common goal, despite to the continuous evolution of the surrounding environment. An important research area for these systems regards the identification of proper load balancing mechanisms that, depending on the current utilization of resources at a node and on its knowledge of the neighbourhood, aim at optimizing at runtime the global system state with simple local actions without a centralized intelligence.

The SelfLet environment is a framework that provides an architectural model and a runtime infrastructure to support the development of distributed and autonomic systems. In this paper we extend the SelfLet approach by defining two optimization policies that, based on a prediction of the future load of a SelfLet node and of its neighbours, compute the most profitable autonomic load balancing action to be actuated. We show that adopting this approach, a system-wide load balancing behaviour emerges from the local actions.

Software in the near ubiquitous future will need to cope with variability, as software systems get deployed on an increasingly large diversity of computing platforms and operates in different execution environments. Heterogeneity of the underlying communication and computing infrastructure, mobility inducing changes to the execution environments and therefore changes to the availability of resources and continuously evolving requirements require software systems to be adaptable according to the context changes. Software systems should also be reliable and meet the user’s requirements and needs. Moreover, due to its pervasiveness, software systems must be dependable. Supporting the validation of these self-adaptive systems to ensure dependability requires a complete rethinking of the software life cycle. The traditional division among static analysis and dynamic analysis is blurred by the need to validate dynamic systems adaptation. Models play a key role in the validation of dependable systems, dynamic adaptation calls for the use of such models at run time. In this paper we describe the approach we have undertaken in recent projects to address the challenge of assessing dependability for adaptive software systems.

A new era is dawning where vast amount of data is subjected to intensive analysis in a cloud computing environment. Over the years, data about a myriad of things, ranging from user clicks to galaxies, have been accumulated, and continue to be collected, on storage media. The increasing availability of such data, along with the abundant supply of compute power and the urge to create useful knowledge, gave rise to a new data analytics paradigm in which data is subjected to intensive analysis, and additional data is created in the process. Meanwhile, a new cloud computing environment has emerged where seemingly limitless compute and storage resources are being provided to host computation and data for multiple users through virtualization technologies. Such a cloud environment is becoming the home for data analytics. Consequently, providing good performance at run-time to data analytics workload is an important issue for cloud management. In this paper, we provide an overview of the data analytics and cloud environment landscapes, and investigate the performance management issues related to running data analytics in the cloud. In particular, we focus on topics such as workload characterization, profiling analytics applications and their pattern of data usage, cloud resource allocation, placement of computation and data and their dynamic migration in the cloud, and performance prediction. In solving such management problems one relies on various run-time analytic models. We discuss approaches for modeling and optimizing the dynamic data analytics workload in the cloud environment. All along, we use the Map-Reduce paradigm as an illustration of data analytics.