應用於匯流排矩陣系統之仲裁器權重調整演算法

Abstract

在系統單晶片匯流排上，仲裁器是必要的元件。當不同裝置同時要求使用匯流排時，會有存取衝突發生，而仲裁器是為了解決這些衝突而存在的。過去一些以樂透方式為基礎的仲裁器演算法是用機率方式去解決這些衝突而且被證明非常的有效。但這些以樂透方式為基礎的仲裁器演算法需要一個權重調整演算法來幫助他們去同時滿足不同裝置的即時以及頻寬的需求。在本篇論文中，我們提出了一個新的權重調整演算法，我們稱作MC權重調整演算法。MC權重調整演算法可以同時考慮到多個匯流排系統的資訊來調整每個裝置的權重。由實驗數據可以證實，我們提出的MC權重調整演算法可以有效的幫助這些以樂透方式為基礎的仲裁器演算法，用以滿足匯流排矩陣上不同裝置的頻寬需求。

Arbiters are mandatory components on SoC bus systems to resolve contentions of bus access requests from different IP cores. Lottery-based arbitration algorithms are probabilistic and efficient arbitration algorithms. However, lottery-based arbitration algorithms need a weight tuning mechanism to help them simultaneously meet both the real-time and bandwidth requirements. In this thesis, we propose a new weight tuning algorithm, named MC weight tuning algorithm, which considers multiple buses at one time. The experimental results show that MC weight tuning algorithm helps lottery-based arbitration algorithms efficiently meet bandwidth requirements of IP cores in bus matrix systems.