Buffered crossbar switch is becoming attractive due to its feature of distributed scheduling, which makes the scheduler more scalable than that in an un-buffered crossbar switch. However, previous researches on buffered crossbar switches mainly aimed at improving the throughput and cell delay performance, scarcely at the fairness and bandwidth guarantee. In this paper, we discuss the necessities of providing bandwidth guarantee in buffered crossbar switches, and propose a novel RR_MCF algorithm. RR_MCF uses a simple round robin policy at input scheduling, and a most credit first policy at output scheduling. By given a predetermined reserved bandwidth matrix, simulations show RR_MCF could provide good bandwidth guarantee for each flow, even with the presence of ill-behaved flows.
To provide QoS control for real time traffic in core routers, this paper designs and evaluates a 320 Gb/s switch system, which supports 16 line cards, each operating at OC192c line rate (10 Gb/s). This switch system contains a high performance switch fabric and supports variable-length IP packet interface. These two characters provide advantages over traditional switch fabrics with a cell interface. This switch system supports eight priorities to both unicast and multicast traffic. The highest priority with strict QoS guarantee is for real time traffic, and other seven lower priorities with weighted round-robin (WRR) service discipline are for other common data traffic. Through simulation under multi-priority burst traffic model, we demonstrate this switch system not only can provide excellent performance for real time traffic, but also can efficiently allocate bandwidth among all kinds of traffic. As a result, this switch system can serve as a key node in high-speed networks, and it can also meet the challenge of multimedia traffic to the next generation Internet.