Unai Arronategui

The study of real discrete event systems requires the use of models to cope with complexity and large scale. The only way to understand and analyse their behaviour prior to implementation is, in practice, through distributed simulation. Although it is a widely studied discipline, the difficulty of developing efficient distributed simulation code remains a challenge. The use of model driven engineering approaches allows a smooth way from informal specifications to executable code showing traces of the system behaviour. Formal models allow to conduct the phases of this engineering process, and in this work, the formalism is Petri nets. In the simulation literature, Petri nets have been shown to be particularly suitable for modelling and simulation of discrete event systems. This article reviews the role of Petri nets as the core formalism to support a model‐driven engineering approach for the execution of large scale models using distributed simulation. It deals with different aspects related to the Petri net‐based languages used at different stages of the modelling and simulation process, from conceptual modelling of complex systems to the generation of code for executing simulations of Petri net‐based models. After the review, the article proposes an efficient representation of Petri net‐based models. It is analysed from the perspective of the essential properties required for distributed simulation, and was found to provide efficient execution, scalability and dynamic configuration. The article highlights the importance of considering modelling constraints in order to guarantee good properties such as liveness and structural boundedness of Petri net components for the execution of large‐scale Petri net models. The Petri net‐based methodology is illustrated from the perspective of the impact of the formalism to help developing well‐formed models and efficient code for distributed simulation.

A key challenge for supporting elastic behaviour in cloud systems is to achieve a good performance in automated (de-)provisioning and scheduling of computing resources. One of the key aspects that can be significant is the overheads associated with deploying, terminating and maintaining resources. Therefore, due to their lower start up and termination overhead, containers are rapidly replacing Virtual Machines (VMs) in many cloud deployments, as the computation instance of choice. In this paper, we analyse the performance of Kubernetes achieved through a Petri net-based performance model. Kubernetes is a container management system for a distributed cluster environment. Our model can be characterised using data from a Kubernetes deployment, and can be exploited for supporting capacity planning and designing Kubernetes-based elastic applications.

Performance interference can occur when various services are executed over the same physical infrastructure in a cloud system. This can lead to performance degradation compared to the execution of services in isolation. This work proposes a Confirmatory Factor Analysis (CFA)-based model to estimate performance interference across containers, caused by the use of CPU, memory and IO across a number of co-hosted applications. The approach provides resource characterization through human comprehensible indices expressed as time series, so the interference in the entire execution lifetime of a service can be analyzed. Our experiments, based on the combination of real services with different profiles executed in Docker containers, suggest that our model can accurately predict the overall execution time, for different service combinations. The approach can be used by a service designer to identify phases, during the execution life-cycle of a service, that are likely to lead to a greater degree of interference, and to ensure that only complementary services are hosted on the same physical machine. Interference-awareness of this kind will enable more intelligent resource management and scheduling for cloud systems, and may be used to dynamically modify scheduling decisions.