Enabling Congestion Avoidance Mechanism in Input Queued Switch Architecture by Eliminating HOL Blocking

Abstract

The evolution of interconnection network represents the dependency of the total system cost and power consumption. To achieve a better performance in High-speed network design, the number of components or resources must be reduced. This leads to a necessity of efficient congestion management technique. RECN (an efficient Head-of-Line blocking elimination technique) is a cost effective Switching architecture to face the challenges of congestion management, it has been recently proposed for Advanced Switching (AS). RECN detects the formation of congestion trees, dynamically allocates queues for storing congested packets, and thus eliminates the HOL blocking introduced by congestion trees. These queues are deallocated when congestion vanishes. In this paper, an enhanced RECN version, specifically distributed queue deallocation mechanism for RECN-IQ (Input queue switch architecture: only queues at input port of a switch) that reduces the number of required resources (Queues), reduces the memory requirement and does not require the use of control packets for the deallocation of SAQ’s (Set Aside Queues- dynamically assigned for congested points) are proposed. Here a new congestion notification mechanism, where flow control packets are used to notify congestion, is used. This simplifies the implementation of RECN-IQ deallocation strategies in AS. Regarding the performance it leads to a significant reduction of the data memory area required at each port in the reduction factor of 5 times than RECN-CIOQ (Combined Input Output Queue - have queues at both Input port & Output port of a switch) and avoids the use of explicit congestion notifications and token-exchanging packets .