Nikolaus Huber

Self-adaptive workload classification and forecasting for proactive resource provisioning

As modern enterprise software systems become increasingly dynamic, workload forecasting techniques are gaining in importance as a foundation for online capacity planning and resource management. Time series analysis offers a broad spectrum of methods to calculate workload forecasts based on history monitoring data. Related work in the field of workload forecasting mostly concentrates on evaluating specific methods and their individual optimisation potential or on predicting Quality-of-Service (QoS) metrics directly. As a basis, we present a survey on established forecasting methods of the time series analysis concerning their benefits and drawbacks and group them according to their computational overheads. In this paper, we propose a novel self-adaptive approach that selects suitable forecasting methods for a given context based on a decision tree and direct feedback cycles together with a corresponding implementation. The user needs to provide only his general forecasting objectives. In several experiments and case studies based on real-world workload traces, we show that our implementation of the approach provides continuous and reliable forecast results at run-time. The results of this extensive evaluation show that the relative error of the individual forecast points is significantly reduced compared to statically applied forecasting methods, e.g. in an exemplary scenario on average by 37%. In a case study, between 55% and 75% of the violations of a given service level agreement can be prevented by applying proactive resource provisioning based on the forecast results of our implementation.

Towards Self-Aware Performance and Resource Management in Modern Service-Oriented Systems

Modern service-oriented systems have increasingly complex loosely-coupled architectures that often exhibit poor performance and resource efficiency and have high operating costs. This is due to the inability to predict at run-time the effect of dynamic changes in the system environment (e.g., varying service workloads) and adapt the system configuration accordingly. In this paper, we describe a long-term vision and approach for designing systems with built-in self-aware performance and resource management capabilities. We advocate the use of architecture-level performance models extracted dynamically from the evolving system configuration and maintained automatically during operation. The models will be exploited at run-time to adapt the system to changes in the environment ensuring that resources are utilized efficiently and performance requirements are continuously satisfied.

Automated extraction of architecture-level performance models of distributed component-based systems

Modern enterprise applications have to satisfy increasingly stringent Quality-of-Service requirements. To ensure that a system meets its performance requirements, the ability to predict its performance under different configurations and workloads is essential. Architecture-level performance models describe performance-relevant aspects of software architectures and execution environments allowing to evaluate different usage profiles as well as system deployment and configuration options. However, building performance models manually requires a lot of time and effort. In this paper, we present a novel automated method for the extraction of architecture-level performance models of distributed component-based systems, based on monitoring data collected at run-time. The method is validated in a case study with the industry-standard SPECjEnterprise2010 Enterprise Java benchmark, a representative software system executed in a realistic environment. The obtained performance predictions match the measurements on the real system within an error margin of mostly 10–20 percent.

Model-based self-adaptive resource allocation in virtualized environments

The adoption of virtualization and Cloud Computing technologies promises a number of benefits such as increased flexibility, better energy efficiency and lower operating costs for IT systems. However, highly variable workloads make it challenging to provide quality-of-service guarantees while at the same time ensuring efficient resource utilization. To avoid violations of service-level agreements (SLAs) or inefficient resource usage, resource allocations have to be adapted continuously during operation to reflect changes in application workloads. In this paper, we present a novel approach to self-adaptive resource allocation in virtualized environments based on online architecture-level performance models. We present a detailed case study of a representative enterprise application, the new SPECjEnterprise2010 benchmark, deployed in a virtualized cluster environment. The case study serves as a proof-of-concept demonstrating the effectiveness and practical applicability of our approach.

Modeling run-time adaptation at the system architecture level in dynamic service-oriented environments

Today, software systems are more and more executed in dynamic, virtualized environments. These environments host diverse applications of different parties, sharing the underlying resources. The goal of this resource sharing is to utilize resources efficiently while ensuring that quality-of-service requirements are continuously satisfied. In such scenarios, complex adaptations to changes in the system environment are still largely performed manually by humans. Over the past decade, autonomic self-adaptation techniques aiming to minimize human intervention have become increasingly popular. However, given that adaptation processes are usually highly system-specific, it is a challenge to abstract from system details, enabling the reuse of adaptation strategies. In this paper, we present S/T/A, a modeling language to describe system adaptation processes at the system architecture level in a generic, human-understandable and reusable way. We apply our approach to multiple different realistic contexts (dynamic resource allocation, run-time adaptation planning, etc.). The results show how a holistic model-based approach can close the gap between complex manual adaptations and their autonomous execution.

Performance modeling in industry: a case study on storage virtualization

In software engineering, performance and the integration of performance analysis methodologies gain increasing importance, especially for complex systems. Well-developed methods and tools can predict non-functional performance properties like response time or resource utilization in early design stages, thus promising time and cost savings. However, as performance modeling and performance prediction is still a young research area, the methods are not yet well-established and in wide-spread industrial use. This work is a case study of the applicability of the Palladio Component Model as a performance prediction method in an industrial environment. We model and analyze different design alternatives for storage virtualization on an IBM* system. The model calibration, validation and evaluation is based on data measured on a System z9* as a proof of concept. The results show that performance predictions can identify performance bottlenecks and evaluate design alternatives in early stages of system development. The experiences gained were that performance modeling helps to understand and analyze a system. Hence, this case study substantiates that performance modeling is applicable in industry and a valuable method for evaluating design decisions.

Modeling dynamic virtualized resource landscapes

Modern data centers are subject to an increasing demand for flexibility. Increased flexibility and dynamics, however, also result in a higher system complexity. This complexity carries on to run-time resource management for Quality-of-Service (QoS) enforcement, rendering design-time approaches for QoS assurance inadequate. In this paper, we present a set of novel meta-models that can be used to describe the resource landscape, the architecture and resource layers of dynamic virtualized data center infrastructures, as well as their run-time adaptation and resource management aspects. With these meta-models we introduce new modeling concepts to improve model-based run-time QoS assurance. We evaluate our meta-models by modeling a representative virtualized service infrastructure and using these model instances for run-time resource allocation. The results demonstrate the benefits of the new meta-models and show how they can be used to improve model-based system adaptation and run-time resource management in dynamic virtualized data centers.

S/T/A: Meta-Modeling Run-Time Adaptation in Component-Based System Architectures

Modern virtualized system environments usually host diverse applications of different parties and aim at utilizing resources efficiently while ensuring that quality-of-service requirements are continuously satisfied. In such scenarios, complex adaptations to changes in the system environment are still largely performed manually by humans. Over the past decade, autonomic self-adaptation techniques aiming to minimize human intervention have become increasingly popular. However, given that adaptation processes are usually highly system specific, it is a challenge to abstract from system details enabling the reuse of adaptation strategies. In this paper, we propose a novel modeling language (meta-model) providing means to describe system adaptation processes at the system architecture level in a generic, human-understandable and reusable way. We apply our approach to three different realistic contexts (dynamic resource allocation, software architecture optimization, and run-time adaptation planning) showing how the gap between complex manual adaptations and their autonomous execution can be closed by using a holistic model-based approach.

Analysis of the performance-influencing factors of virtualization platforms

Nowadays, virtualization solutions are gaining increasing importance. By enabling the sharing of physical resources, thus making resource usage more efficient, they promise energy and cost savings. Additionally, virtualization is the key enabling technology for Cloud Computing and server consolidation. However, the effects of sharing resources on system performance are not yet well-understood. This makes performance prediction and performance management of services deployed in such dynamic systems very challenging. Because of the large variety of virtualization solutions, a generic approach to predict the performance influences of virtualization platforms is highly desirable. In this paper, we present a hierarchical model capturing the major performance-relevant factors of virtualization platforms. We then propose a general methodology to quantify the influence of the identified factors based on an empirical approach using benchmarks. Finally, we present a case study of Citrix XenServer 5.5, a state-of-the-art virtualization platform.

A Method for Experimental Analysis and Modeling of Virtualization Performance Overhead

Nowadays, virtualization solutions are gaining increasing importance. By enabling the sharing of physical resources, thus making resource usage more efficient, they promise energy and cost savings. Additionally, virtualization is the key enabling technology for cloud computing and server consolidation. However, resource sharing and other factors have direct effects on system performance, which are not yet well-understood. Hence, performance prediction and performance management of services deployed in virtualized environments like public and private clouds is a challenging task. Because of the large variety of virtualization solutions, a generic approach to predict the performance overhead of services running on virtualization platforms is highly desirable. In this paper, we present a methodology to quantify the influence of the identified performance-relevant factors based on an empirical approach using benchmarks. We show experimental results on two popular state-of-the-art virtualization platforms, Citrix XenServer 5.5 and VMware ESX 4.0, as representatives of the two major hypervisor architectures. Based on these results, we propose a basic, generic performance prediction model for the two different types of hypervisor architectures. The target is to predict the performance overhead for executing services on virtualized platforms.