An Efficient Routing Strategy for Networks-on-Chip with the Dynamic Architecture of the Switch
Date Issued
2007
Date
2007
Author(s)
Huang, Min-Xuan
DOI
en-US
Abstract
我們提出和評估一個新穎的路徑技術，這個技術結合了決定行演算法和完全適應型演算法。我採用了XY路徑演算法和奇偶數轉向演算法的優點。更準確的說法，是我們展望了一個可以明智而審慎地在XY路徑演算法和奇偶數轉向演算法轉換的新技術，而這樣的轉換是根據網路的壅塞情況來決定的。在模擬的結果中，可以見到所提出的動態路徑演算法結合XY路徑演算法和奇偶數轉向演算法的效能在和純粹的XY路徑演算法和奇偶數轉向演算法，而這樣的模擬是根據不同的交通型態下進行。因此，一個標準路由器以動態路徑演算法結合XY路徑演算法和奇偶數轉向演算法為主被設計和評估。相較於純粹的適應型演算法來說，實做一個動態路徑演算法結合XY路徑演算法和奇偶數轉向演算法的酬載可以被忽略的。其實，在兩者的相對比較下，奇偶數轉向演算法的酬載是動態路徑演算法結合XY路徑演算法和奇偶數轉向演算法的0.4%。然而，動態路徑演算法結合XY路徑演算法和奇偶數轉向演算法的性能固然地比XY路徑演算法和奇偶數轉向演算法還要好。而其中，動態路徑演算法的性能可以比奇偶數轉向演算法好上13.1%。
We present and evaluate a novel routing scheme which combines the deterministic routing algorithm and the full adaptive routing algorithm. We combine the advantages of both xy routing algorithm and odd-even turn model. More precisely, we envision a new routing technique which judiciously switches between deterministic and adaptive routing based on the network’s congestion conditions. The simulation results show the effectiveness of the Dynamic routing algorithm combining the deterministic and adaptive model by comparing it with pure xy routing algorithm and odd-even turn model under different traffic patterns. Moreover, a prototype router based on the Dynamic routing algorithm combining the deterministic and adaptive model has been designed and evaluated. Compared to purely adaptive routers, the overhead of implementing Dynamic routing algorithm combining the deterministic and adaptive model is negligible. The overhead of the odd-even turn model is 0.4% higher compared to the overhead of the dynamic routing algorithm. However, the performance is consistently better than the one of the xy routing and the one of the odd-even turn model. The performance is 13.1% better compared to the odd-even turn model.
Subjects
晶片網路
網狀物
路徑演算法
低功率
Network-on-Chip
Mesh
routing algorithm
Low power
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-96-R94922156-1.pdf
Size
23.31 KB
Format
Adobe PDF
Checksum
(MD5):6e2d58a2fab5d6f18166c534c3685f2e