光互连网络中资源预留的多级光互连仲裁机制

基于资源预留策略提出一种多级光互连仲裁机制,通过将网络分级实现快速、高效的仲裁。多优先级数据缓存队列的传输节点设计,提供了不同类型流量的差异化传输;通过预约式两级仲裁机制,实现网络的完全公平与100%的高吞吐率。设计并对快速仲裁通道进行了合理布局,极大地缩短了仲裁延迟。仿真结果表明:采用基于资源预留的分级仲裁策略,在多种流量模式下所有节点均获得公平的服务。与Feather Weight相比,分级仲裁策略吞吐率提高17%;与2-pass相比,仲裁延迟减少15%,同时,功耗减少5%。

A Multiple-Level Arbitration in Optical Network-on-Chip based on Resource Reservation

we propose a multi-level arbitration based on resource reservation in optical network-on-chip. The fast and high efficiency arbitration scheme divides the network into multi-stages. With the design of multi priority queues in every node, arbiters provide differential transmissions for different kinds of flows. The arbiter also presents the transmit bund resource reservation scheme to fairly reserve time slots for all nodes, and therefore achieving the throughput of 100%. Propose and layout the fast arbitration channels (FACs) to degrade the arbitration period, reducing the packet transmitting delay. The simulation results show that, with the multi-level arbitration based resource reservation, all nodes are allocated with almost equal service under various patterns. This scheme improves throughput by 17% when comparing to FeatherWeight under the self-similar traffic pattern and decreases arbitration delay by 15% to 2-pass arbitration, incurring total power overhead of 5%.