Gabriel Iszlai

Resource provisioning for cloud computing

In resource provisioning for cloud computing, an important issue is how resources may be allocated to an application mix such that the service level agreements (SLAs) of all applications are met. A performance model with two interactive job classes is used to determine the smallest number of servers required to meet the SLAs of both classes. For each class, the SLA is specified by the relationship: Prob [response time ≤ x] ≥ y. Two server allocation strategies are considered: shared allocation (SA) and dedicated allocation (DA). For the case of FCFS scheduling, analytic results for response time distribution are used to develop a heuristic algorithm that determines an allocation strategy (SA or DA) that requires the smallest number of servers. The effectiveness of this algorithm is evaluated over a range of operating conditions. The performance of SA with non-FCFS scheduling is also investigated. Among the scheduling disciplines considered, a new discipline called probability dependent priority is found to have the best performance in terms of requiring the smallest number of servers.

Performance model driven QoS guarantees and optimization in clouds

This paper presents a method for achieving optimization in clouds by using performance models in the development, deployment and operations of the applications running in the cloud. We show the architecture of the cloud, the services offered by the cloud to support optimization and the methodology used by developers to enable runtime optimization of the clouds. An optimization algorithm is presented which accommodates different goals, different scopes and timescales of optimization actions, and different control algorithms. The optimization here maximizes profits in the cloud constrained by QoS and SLAs across a large variety of workloads.

Exploring Alternative Approaches to Implement an Elasticity Policy

An elasticity policy governs how and when resources (e.g., application server instances at the PaaS layer) are added to and/or removed from a cloud environment. The elasticity policy can be implemented as a conventional control loop or as a set of heuristic rules. In the control-theoretic approach, complex constructs such as tracking filters, estimators, regulators, and controllers are utilized. In the heuristic, rule-based approach, various alerts(e.g., events) are defined on instance metrics (e.g., CPU utilization), which are then aggregated at a global scale in order to make provisioning decisions for a given application tier. This work provides an overview of our experiences designing and working with both approaches to construct an auto scaler for simple applications. We enumerate different criteria such as design complexity, ease of comprehension, and maintenance upon which we form an informal comparison between the different methods. We conclude with a brief discussion of how these approaches can be used in the governance of resources to better meet a high-level goal over time.

A business driven cloud optimization architecture

In this paper, we discuss several facets of optimization in cloud computing, the corresponding challenges and propose an architecture for addressing those challenges. We consider a layered cloud where various cloud layers virtualize parts of the cloud infrastructure. The architecture takes into account different stakeholders in the cloud (infrastructure providers, platform providers, application providers and end users). The architecture supports self-management by automating most of the activities pertaining to optimization: monitoring, analysis and prediction, planning and execution.

Optimal autoscaling in a IaaS cloud

An application provider leases resources (i.e., virtual machine instances) of variable configurations from a IaaS provider over some lease duration (typically one hour). The application provider (i.e., consumer) would like to minimize their cost while meeting all service level obligations (SLOs). The mechanism of adding and removing resources at runtime is referred to as autoscaling. The process of autoscaling is automated through the use of a management component referred to as an autoscaler. This paper introduces a novel autoscaling approach in which both cloud and application dynamics are modeled in the context of a stochastic, model predictive control problem. The approach exploits trade-off between satisfying performance related objectives for the consumers application while minimizing their cost. Simulation results are presented demonstrating the efficacy of this new approach.

Feedback-based optimization of a private cloud

The optimization problem addressed by this paper involves the allocation of resources in a private cloud such that cost to the provider is minimized (through the maximization of resource sharing) while attempting to meet all client application requirements (as specified in the SLAs). At the heart of any optimization based resource allocation algorithm, there are two models: one that relates the application level quality of service to the given set of resources and one that maps a given service level and resource consumption to profit metrics. In this paper we investigate the optimization loop in which each applications performance model is dynamically updated at runtime to adapt to the changes in the system. These changes could be perturbations in the environment that had not been included in the model. Through experimentation we show that using these tracking models in the optimization loop will result in a more accurate optimization and thus result in the generation of greater profit.

Tracking adaptive performance models using dynamic clustering of user classes

Estimation techniques have been largely applied to track hidden performance parameters (e.g. service demands) of web based software systems. In this paper we investigate dynamic multiclass modeling of such systems, with variable classes of service, aiming at finding a low complexity model yet with enough accuracy. We propose a combination of clustering algorithm and tracking filter for effective grouping of classes of services. The tracking estimator is based on a layered queuing model with parameters for CPU demands and the user load intensity of each class of service. Clustering uses the K-means algorithm. The target application is autonomic control of web clusters, where changes occur at different rates and amplitudes and at random time instants. Experiments show that the tracking is effective, and reveal good filter settings for different variations.

Designing autonomic management systems for cloud computing

Autonomic Computing Systems are systems which are capable of adapting themselves to changes in their working environment in order to maintain required service level agreements, protect the execution of the system from external attacks or prevent and recover from failures. Within the field of autonomic computing, autonomic systems are developed as control loops which monitor and analyze the execution of the system and then plan and execute changes if needed in order to adapt the system to its environment. This paper will present an approach for designing and building autonomic systems for cloud computing, based on an architecture previously developed which was rooted in real-time software patters. Furthermore, the paper presents an application of the autonomic management architecture to a cluster of application servers running on top of a cloud. It is thus demonstrated how the development approach can be easily reconfigured for the control and supervision of different types of autonomic computing strategies such as self-management for Web Services or self-provisioning and self-optimization for server virtualization.

Capacity planning for service-oriented architectures

Service-oriented architectures (SOAs) are being increasingly adopted for the development of distributed applications that involve multiple partner organizations. The main challenge in configuring such applications - whether autonomously or manually - is meeting the service quality expected by the consumers. In this paper, we describe a methodology and corresponding tool implementation for estimating the capacity of alternative configurations of complex service-oriented applications. We use a sophisticated enterprise application with many possible configurations as our test application. The current tool prototype simulates the behavior of the application for a given configuration on an existing network topology. This simulation is relatively coarse-grained, but is capable of tracking several performance indicators. We evaluate this simulation output against actual performance data.

Observability and controllability of autonomic computing systems for composed Web services

Autonomic Computing is a research area whose aim is to embed “intelligent algorithms” in the IT infrastructure management software such that it can adapt to changes in regards to the configuration, provisioning, external attacks, and resource utilization variations at run time. It is therefore, almost natural to consider this IT infrastructure control software framework as being designed upon methods and technologies used for the design of control systems. In this paper the control system design methodology is extended to the analysis of the intrinsic properties of the autonomic system itself. Thus the controllability and observability properties of the computing process itself are defined and examined in more details. These properties are also investigated for the case of cloud services where the serial and parallel composition of these services is considered. These cloud based services are connected through cooperation protocols that define a global process dynamic. Web services are modeled as scheduled computational processes waiting in a queue to cooperate in delivering the service. This paper proposes an input-state-output mathematical model for the autonomic computing model of cloud based services and the observability and controllability are further analyzed on the above models. As an example a Kalman based control is applied to such processes and the general architecture and some simulation results are given.