Nabila Salmi

An Approach for Performance Modelling and Analysis of Multi-tiers Autonomic Systems

The need of a modern system to be more autonomous with regard to its environment make autonomic systems very attractive, providing them with self-management capabilities. Usually, an autonomic system reconfigures itself to overcome some problem or optimize itself. However, these reconfigurations may result in a loss of performance or even a service degradation. To avoid such situations, this paper proposes a performance prediction approach for autonomic systems, based on formal modelling. Our goal is to forecast the most appropriate configuration for an autonomic system, particularly, a multi-tier system, in order to get an efficient use of resources. The main idea is to model resource allocation triggered by the autonomic loop for achieving the best configuration. Then, we show how to analyse the whole system and evaluate the autonomic loop impact on system performances. The method is based on Stochastic Petri Nets (SPN) modelling. The resource availability, the workload intensity and the autonomic features are targeted. A typical example of a multiers system with one server is presented, analysed with the GreatSPN tool, to show the effectiveness of the proposed approach.

Towards Efficient Component Performance Analysis in Component Based Architectures

The desire to bring better quality and higher efficiency in software design has led to the development of Component Based Systems. This kind of development has several benefits, however, at the performance level, no guarantees ensure software correctness and good performance properties. To help application designers to meet desired performance of their applications, this paper proposes a modular analysis process that allows to assess independently and efficiently component performances and its impact on a component based architecture. This process is achieved through a modelling phase, based on Stochastic Well-formed Nets (SWN), a high level model of Stochastic Petri nets, and a compositional structured performance evaluation method. It starts from the system definition given in a suitable Architecture Description Language, the targeted component implementation and an ”abstract view” of other components, then provides efficiently system performance indexes. The process is illustrated through an application example.

Structured performance analysis for component-based systems

The component-based system (CBS) paradigm is now largely used to design software systems. In addition, performance and behavioural analysis remains a required step for the design and the construction of efficient systems. This is especially the case of CBS, which involve interconnected components running concurrent processes. This paper proposes a compositional method for modelling and structured performance analysis of CBS. Modelling is based on stochastic well-formed nets (SWNs), a high level model of stochastic Petri nets, widely used for dependability analysis of concurrent systems. Starting from the definition of the system given in a suitable architecture description language, and from the definition of the elementary components, we build an SWN of the global system together with a set of SWNs modelling the components of the CBS and their connections. From these models, we derive performances of the system thanks to a structured analysis induced by the structure of the CBS. We describe the application of our method through an example designed in the framework of the CORBA component model.

Performance evaluation of Fractal component-based systems

Component-based system development is now a well accepted design approach in software engineering. Numerous component models have been proposed, and for most of them, specific software tools allow building component-based systems (CBS). Although these tools perform several checks on the built system, few of them provide formal verification of behavioural properties nor performance evaluation of the resulting system. In this context, we have developed a general method associating to a CBS, a formal model, based on stochastic well formed nets, a class of high-level Petri nets, allowing qualitative behavioural analysis together with performance evaluation of this CBS. The definition of the model heavily depends on the (run time) component model used to describe the CBS. In this paper, we instantiate our method to Fractal CBS and its reference Java implementation Julia. The method starts from the Fractal architectural description of a system and defines rules to systematically generate element models of the CBS and their interactions. We then apply a structured method for both qualitative and performance analysis, taking into account the given implementation of the Fractal model. The main interest of our method is to take advantage of the compositional definition of such systems to carry out an efficient analysis. The paper concentrates on performance evaluation and presents our method step by step with an illustrative example.

Achieving Scalability of Self-optimizing Multi-tier Systems Performance Prediction

To achieve better responsiveness in nowadays computing systems, autonomic systems have been developed. Such systems are enriched with self-management properties. In particular, self-optimizing systems are autonomic systems which optimize their configuration automatically, to reach a successful QoS. This is particularly useful for modern cloud and internet multi-tier systems. However, introducing such properties may lead to intensive use of system resources, and so to low system operability and poor performances. To check and avoid such situations, we want to provide a global approach for performance modelling of autonomic multi-tier, allowing to forecast their behaviour and efficiency. A first proposal has been introduced in our previous work for multi-tiers self-optimizing systems, to model a system with a Stochastic Petri Net (SPN) model in an incremental manner. The approach was experimented on a one-tier system. Even though, dealing with several tiers often results in a huge model, difficult to analyze within a reasonable time. This paper addresses the growing scalability of our proposed modelling and performance prediction. We generalize our methodology to several tiers and propose a Petri net reduction method to overcome the scalability issue. We illustrate the effectiveness of our approach through a set of experimental analysis results.

Towards Modelling-Based Self-adaptive Resource Allocation in Multi-tiers Cloud Systems

Achieving efficient resource allocation is one of the most challenging problems faced by cloud providers. These providers usually maintain hosted web applications within multiple tiers over the cloud, leading to an overall increased complexity. To answer user requests, meet their Service Level Agreements SLA and reduce the energy cost of the data center, cloud systems are being enforced with self-adaptive features such as self-scaling, to efficiently automate the resource allocation process. However, the main concern is how to choose the best resource configuration to reach these objectives of Quality of Service QoS with a minimal amount of resources consumption. In this context, we target to use performance modelling and analysis, to forecast the system performances and deduce the most appropriate resource configuration to be applied by the autonomic manager. As a first work to define a modelling based resource allocation autonomic manager, we present, in this paper, the modelling and analysis process, allowing to predict the efficiency of the self-adaptive systems relating resource allocation in the context of multi-tiers cloud systems. We used Stochastic Petri Nets modelling, enforced with a reduction method to avoid a scalability issue. A set of experiments illustrates our approach starting from modelling to performance evaluation of the studied system.

Using performance modelling and analysis for self-adaptive resources allocation systems: A case study

Data centers need to have more and more flexible execution environments, allowing resources sharing between their different applications in order to meet performances requirements. In a cloud computing application for instance, the main objective is to maximize profits by an efficient resources use, to meet the clients Service Level Agreements (SLA) and reduce the energy cost of the data center. The main challenge of resource allocation is then to find the minimum amount of resources that an application needs to meet the desired Quality of Service. To answer these concerns, self-management capabilities have been proposed to efficiently automate the resource allocation process. Autonomic managers allow to adjust the scale of the targeted systems, based on a simple monitoring process and predefined scaling strategies. In this context, it becomes important to forecast the efficiency of such self-adaptive systems, so that to find the most appropriate resource configuration to be applied. To reach this objective, we present, in this paper, a modelling approach, allowing to predict the efficiency of self-adaptive systems relating resource allocation. We use, for this purpose, a Stochastic Petri Nets modelling. A set of experiments illustrates our approach starting from modelling to performance evaluation of the studied system.

Accompanying Component Based Systems Dynamic Reconfiguration with Formal Modelling and Analysis

Nowadays systems should be able to perform evolutionary changes without degrading performances of online services. To achieve that, it is interesting to predict performances of a system reconfiguration, as introducing such properties in a system may lead to quality of service loss and performance degradation. In this case, the impact analysis of a reconfiguration before applying it effectively becomes a challenge, to help software engineers in analyzing their applications and deciding whether a reconfiguration should be done or discarded to avoid performance problems. We are interested in that concern, in the field of Component-Based Systems (CBS). In our previous work, we proposed, a new formalism for checking consistency of dynamic reconfigurations of component based systems. In this paper, we provide a new approach for formal Modelling of a dynamic reconfiguration on CBS to allow quantitative analysis. The modelling consists of generating stochastic Well-formed net (SWN) models, starting from reconfiguration description. Performance indices can be computed through the analysis of obtained models with an SWN tool such as the GreatSPN package used here. A case study of a Fractalbased system reconfiguration illustrates the effectiveness of our approach.

Towards Reliability and Performance Prediction of Autonomic Systems with Self-Healing and Protection

Autonomic systems providing self-healing and self-protection capabilities have been proposed to efficiently automate rectification of system faults and recovery from malicious attacks. In fact, it becomes more and more difficult, labor-intensive, expensive and error-prone to conduct such recoveries. Self-healing techniques and security mechanisms are resource intensive and may affect system performances and even its full operability. Therefore, balancing security and performance in these systems is needed and self-management strategies should guarantee a minimal level of functionality. In our work, we are interested to provide methodologies and tools to predict the behaviour and efficiency of autonomic strategies relating self-healing and self-protection, before applying some healing solutions. The idea is to forecast the most appropriate configuration and ensure the effectiveness of the autonomic manager after application of a solution. So, we propose, in this paper, a general modelling methodology of an autonomic system implementing self-healing and protection, based on stochastic Petri nets. We consider in our modelling an autonomic diagnostic and recovery of fault-tolerant multi-tier systems, directed by the workload intensity, possible attacks and failure frequencies. We illustrate the effectiveness of our approach through a set of experimental analysis results.

Using Performance Modelling for Self-Healing and Protection in Autonomic Multi-tier Systems

We propose, in this paper, a general modelling technique of an autonomic multi-tier system implementing self-healing and self-protection properties. The modelling is based on stochastic Petri nets (SPN). Our objective is to forecast the most appropriate configuration for a multi-tier system and ensure the effectiveness of the autonomic manager. We consider in our modelling an autonomic diagnostic and recovery, with fault tolerance solutions. The diagnostic is directed by the workload intensity, possible attacks and failure frequency. Finally, we illustrate the effectiveness of our modelling through a set of experimental analysis results.