Runtime adaptive QOS management in NOC-based MPSOCS

• Ano de defesa: 2014
• Autor(a) principal: Ruaro, Marcelo
• Orientador(a): Moraes, Fernando Gehm
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Porto Alegre
• Palavras-chave em Português: INFORMÁTICA; ARQUITETURA DE COMPUTADOR; MULTIPROCESSADORES
• Link de acesso: http://hdl.handle.net/10923/5828

Multiprocessor systems on chip (MPSoCs), using networks on chip (NoC) as the communication infrastructure, result from the continuous reduction in the transistors size and the need for increasing computational power. This increased computing capacity is obtained through the reuse of components (processors, memories, routers, etc. ), which also provides scalability, and simplifies the design process. MPSoCs with hundreds of processing elements (PEs) follows the Moore's law, and according to the ITRS 2011 it is predicted up to 1000 PEs in a single chip at the end of 2025. This estimation is driven mainly by the telecommunications and multimedia market, which includes devices such as smartphones and mobile computers. Such devices require systems able to execute a wide range of applications, with different performance requirements. Thus, the system must be able to provide quality of service (QoS) to applications, and adjust the resources usage at runtime. Literature proposais provide runtime QoS adaptation taking finto consideration the use of only one or two QoS techniques. Considering the application diversity that may execute in MPSoCs, applications may have different QoS requirements, requiring more computing resources, communication resources, or both simultaneously. Therefore, this work aims to explore different QoS levels addressing four different adaptive QoS techniques managed at runtime according to the soft real-time applications' requirements. It is assumed that the MPSoC is partitioned in regions, named clusters, with one manager PE per cluster. This adaptive management is controlled through a heuristic that is executed by the OS of each cluster manager. A hybrid monitoring infrastructure provides the necessary information for the adaptive heuristic. This infrastructure is divided in two hierarchical levels, being scalable and with an intrusion levei that corresponda, in the worst case, to 0. 8% of the link utilization. The runtime adaptive QoS management acta in computing, communication, or both, enabling soft real time applications to restore their performance after detected a performance decrease by monitoring. Furthermore, a debugging tool for NoC-based MPSoCs is proposed in this work. This tool provides a communication protocol level debugging, and helps the process of implementation, validation and extraction of results of new system protocols.