ABSTRACT Frame-Aggregated Concurrent Matching Switch

Network operators need high-capacity router architectures that can offer scalability, provide throughput and performance guarantees, and maintain packet ordering. However, previous router architectures based on centralized crossbarbased architectures cannot scale to fast line rates and high port counts. Recently, a new scalable router architecture called the Concurrent Matching Switch (CMS) [5] was introduced that offers scalability by utilizing a fully distributed architecture based on two identical stages of fixed configuration meshes. It has been shown that fixed configuration meshes can be scaled to very fast line rates and high port counts via optical implementations. It has also been shown that the CMS architecture can achieve 100 % throughput and packet ordering with only sequential hardware and O(1) amortized time complexity operations at each linecard. However, no delay performance guarantees have been shown for CMS. In this paper, we demonstrate a general delay performance guarantee for CMS. Based on this guarantee, we propose a novel frame-based CMS architecture that can achieve a performance guarantee of O(N log N) average packet delay, where N is the number of switch ports, while retaining scalability, throughput guarantees, packet ordering, and O(1) time complexity. This architecture improves upon the best previously-known average delay bound of O(N 2) given these switch properties. We further introduce several alternative frame-based CMS architectures