Certificação de componentes em uma plataforma de nuvens computacionais para serviços de computação de alto desempenho.

Abstract

The development of correct and safe High Performance Computing (HPC) applications is a challenge for developers, since such applications generally use parallelism and run on heterogeneous parallel computing platforms. The Doctoral Thesis proposed in this document is aimed at presenting an architecture of a component certification mechanism for cloud computing platforms of high performance computing services. In particular, this mechanism is proposed within the context of the HPC Shelf platform, allowing the construction of certified components for functional and non-functional properties, which can be used to compose applications for expert users. Two particular certifier components are proposed using the certification mechanism introduced in this Thesis: SWC2 (Scientific Workflow Certifier Component) e C4 (Computation Component Certifier Component). SWC2 components are used to verify formal properties of workflows in HPC Shelf. In turn, C4 components are employed to verify formal properties on computation components. There are still tactical components, which expose the services of software formal verification infrastructures and can be orchestrated, by certifiers, by means of the TCOL (Tactical Component Orchestration Language) language, also proposed in this work. It is expected to contribute to the state-of-the-art in the following points: in cloud computing, by providing the first cloud infrastructure focused on software formal verification using exclusively high performance computing techniques; in component-oriented platforms, by providing nondisruptive components that can certify others in a reflexive way; enabling the creation of the so-called parallel certification systems, which are formed by the orchestration of provers to verify formal properties; in scientific workflows, by extracting the main verifiable patterns in these workflows; and in high performance computing applications, by providing a study on which software formal verification tools are able to verify their properties.