Rede Omega Virtual em FPGA com reconfiguração em tempo de execução: estudo de caso: cálculo de atratores em redes reguladoras de genes

• Autor(a) principal: Vendramini, Julio Cesar Goldner
• Data de Publicação: 2012
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: LOCUS Repositório Institucional da UFV
• Texto Completo: http://locus.ufv.br/handle/123456789/2634
• topic: FPGA; Rede Omega; Camada virtual; FPGA; Omega Networks; Virtual layer; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

The multistage interconnection networks emerged in 1950 to be used on telephony systems. During the last decades, multistage networks have still been used on many parallel applications. This work analyzes Omega interconnection networks as a communication approach to be used on parallel architectures mapped on FPGAs. The proposed approach is based on a virtual layer above the FPGA s programmation as a logical level. Several topics were covered. First, the behavior of networks has been analyzed as a function of its workload. We show that Omega network is non-blocking under a partial workload. Moreover, we propose to use the radix4 switches which have a better routing capacity. In addition, the radix4 network reduces the area and the latency of an FPGA implementation. Besides, the regularity of the network was captured with FPGA synthesis tools. Third, one of the problems of multistage networks is the time to configure or routing the connections. We present two routing algorithms on hardware that can be used at runtime. One of the algorithms is based on embedded memories that has a low implementation cost. However, the algorithm spends between two and ten cycles to route one connection. The second algorithm uses a priority encoder and has a high cost compared to the first. Although, this algorithm only takes two cycles to route any connection. Finally, two case study are presented to show how a Omega networks is an efficient option as communication mechanism in FPGA based reconfigurable architectures. The first has a global interconnection network to interconnect heterogeneous processing units in a coarse-grained reconfigurable architecture (CGRA). The second uses an interconnection network in a bioinformatics algorithm. The algorithm computes the attractors on gene regulatory networks modeled with Boolean graphs and scale free topology. This algorithm explores an exponential space solution for the calculation of period attractors. The results showed a gain of 2-3 orders of magnitude over the solution implemented in software with general purpose processors.