基于输入端无冲突算法的面向输出排队的交换结构

Output-Oriented Queued Switch Architecture with Input Conflict-Free Algorithm

该文提出了一种全新的面向输出排队的交换结构,该结构将信元存储于输入端,而信元的调度请求面向输出端排队.利用输入端无冲突调度算法,可以使结构对存储器带宽的需求和输入排队交换机一致.该文设计的调度矩阵使算法复杂度和端口规模呈线性关系,且每一步只需要一次按位"与"操作.文章同时证明了结构要达到稳定的充分条件是使用2倍传输加速比.仿真实验表明,对于均匀流量,该结构时延性能和主流交换结构相似,且抖动性能远优于主流交换结构,而在2倍传输加速比时,其时延性能和OQ完全相同;对于非均匀流量,该结构吞吐率性能优于主流交换结构,且在使用1.14倍传输加速比时,其吞吐率性能和OQ相同.

This paper proposes a new packet switching architecture, called output-oriented queued （OOQ） switch. Unlike any existing switching architectures, OOQ stores cells at input units, while the requests for scheduling is queued at output units. With a novel input conflict-free （ICF） algorithm proposed by the authors, the bandwidth requirement of memories in the switch remains the same as in traditional input-queued switches. To support ICF algorithm, a hardwarebased scheduling matrix is also proposed to make the complexity of the algorithm linear as the switch size. And each step of the scheduling consists of only one parallel bitwise ＂and＂ operation. This paper also prove that the sufficient condition for OOQ/ICF to be stable under any admissible traffic： is to use the transfer speedup of 2. Finally, the simulation results show that under uniform traffic, delays of OOQ/ICF without speedup are comparable to the mainstream switches but with much better delay jitters. Furthermore, with the speedup of 2, the delay performance of OOQ/ ICF is exactly the same as the output-queued switch. While under non-uniform traffic, the throughput of OOQ/ICF is better than current using the speedup of 1.14.