The CICQ Switch with Virtual Crosspoint Queues for Large RTT

The memory size required for a combined input and crosspoint queued (CICQ) switch with an existing credit-based flow control is proportional to the flow control latency between the line cards and switch fabric. Recently, scalable packet switches were implemented with distributed units introducing a large RTT latency between the line cards and switch fabric, making the CP buffer requirement impractical. In this paper, a CICQ switch with virtual crosspoint queues (VCQs) is investigated. The VCQs unit resides inside the switch fabric and is dynamically shared among virtual output queues (VOQ) from the same source port and is operated at the line rate, making the implementation practical. The CICQ switch with VCQ can achieve high throughput for unbalanced traffic while reducing over 87% of the total buffer size requirement in the switch fabric, thus making the scalable implementation of a distributed CICQ switch practical.

I. Introduction

A combined input and crosspoint queued (CICQ) switch [13] [14] [18] has been considered to be the next generation of packet switches for its scalability [1]. As shown in Fig. 1, the CICQ switch has buffering at each input port and at each crosspoint (CP) internal to the switch fabric. Credit-based flow control from the switch fabric is used to control the flow of packets from the input ports to the CP buffers. In the CICQ switch, a VOQ scheduler, VOQ-S, at each input port and CP schedulers, CP-Ss, inside the switch fabric can run in parallel and independently [18]. Thus, the synchronization among input port schedulers used for input queued (IQ) switch is no longer needed.