چكيده به لاتين
Abstract:
Congestion issue is the major factor, which limits improvement of network-on-chip efficiency. It leads to larger transmission latency, higher thermal and more power consumption. Congestion-aware routing algorithms, which are classified as local, regional and global algorithms, can distribute traffic load over the network and reduce packet transmission latency by selecting less congested paths. In this thesis, two congestion-aware routing algorithms are presented to distribute traffic load in 3D NoCs.
The first algorithm is a globally congestion-aware routing in which each router is aware about congestion status of all network nodes. This method, which named Globally Congestion Aware Routing Algorithm (G-CARA), propagates congestion data through the network by embedding status information in data packets. After propagation of congestion information, congestion-aware routing is performed in each network layer based on XY, So that data packets reach their destination with the minimum of delay. The second algorithm is a regionally congestion-aware routing in which each router is aware about congestion status of all nods that are located in the stair region. In this method, which named Regional Congestion Aware Routing Algorithm (R-CARA), a low-latency congestion propagation network is implemented for collecting and distributing traffic data. For routing packets, R-CARA introduces four situations according to position of current router and destination. It estimates the congestion of admissible directions based on these situations and forwards packet in less congested direction.
Under different traffic patterns, simulation results show that the first algorithm improves the saturation throughput and average packet latency in comparison with other algorithms, an improvement of 14.68%-22.95% and 42.43%-61.29% respective. The second algorithm improves the saturation throughput and average packet latency in comparison with existing congestion-aware schemes, an improvement of 5.04%-17.08% and 18.62%-51.21% respective.
Keywords: 3D Network-on-Chip (NoC), Adaptive Routing, Congestion-Aware, Deadlock freedom.
