Conillon: Distributed computing platform for desktop grids

• Autor(a) principal: Silva, Hélio Alexandre Dias da
• Data de Publicação: 2011
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/8398
• topic: Distributed computing; Parallel computing; High performance computing; Fault tolerance; Performance; Java; Computação distribuída; Computação paralela; Computação de alta performance; Tolerância a falhas

Grid computing enables organizations to integrate and share sets of heterogeneous resources into one synergetic and powerful system, in order to deliver e ciency and processing performance to demanding parallel applications. The aggregation and the coordination of these resources need to be integrated in a set of tools and protocols that reduce their complexity. Resource allocation is performed by a centralized manager that has complete knowledge of system state. Conillon is a distributed computing platform that manages and coordinates a set of actors. This platform enables heterogeneous workstations to function as part of a distributed system. Conillon is a lightweight, simple to use, and yet allows any number of di erent applications to be deployed and executed in parallel. Our platform maximizes the resource utilization in bene t of the demanding parallel applications. Parallel applications developers have access to a set of tools that simpli es the development process and the creation of tasks. The complexity related to the scheduling, distribution, and fault tolerant management of tasks is shielded from the programmer. Conillom platform has an integrated set of programms that work synergistically delivering a positivo ratio between e ciency and processing performance. We present our platform in detail as well as the results of extensive experiments about perfor- mance. The platform behaved relatively well and achieves an average speed up of up to x1.9 on the time needed to complete a job each time the computational resources are doubled.