节点文献

交叉节点带缓存的Crossbar交换结构及调度算法研究

Research on the Buffered Crossbar Switching Fabrics and Its Scheduling Algorithm

【作者】 王晓亮

【导师】 周代琪; 邱智亮;

【作者基本信息】 西安电子科技大学 , 通信与信息系统, 2006, 硕士

【摘要】 伴随着Internet网络迅速发展,特别是多媒体和数据业务的爆炸式增长,互联网的数据流量急剧增加。骨干网需要高速,大容量的T比特级路由交换设备。高性能的交换网络是路由器的核心部件,对路由器的性能有重要影响。而交换网络的调度算法主要实现网络输入端口和输出端口之间的匹配,从而解决分组对网络资源的竞争,达到合理利用网络资源,提高吞吐率,减少时延的目的。调度算法的好坏会直接影响整个交换网络的性能,从而影响路由器的整体性能。所以对交换网络及其调度算法的研究是具有重要意义的。本文的研究重点集中在一种交叉节点带缓存的Crossbar结构。由于硬件工艺水平的发展,在芯片内部实现大量缓存成为可能,交叉节点带缓存的Crossbar结构因其简单的调度,良好的性能逐渐成为近年来交换网络方面的研究热点。首先,文章分析了采用组合输入交叉节排队的CICQ的结构特性,介绍了CICQ交换结构的研究背景,对现有的研究成果从不同角度进行分类。并在各种输入业务流条件下,采用仿真方法对影响CICQ交换结构性能的交叉点缓存大小,往返时延,加速比和端口数等因素进行了详细的分析。其次,对CICQ结构调度算法进行研究。对已有算法做了简单的总结和分析。根据IP分组的不同交换方式,分别从定长信元方式交换和变长分组方式交换两个方面进行讨论。提出无权重的静态指针轮询算法—APRR算法。该算法实现简单,执行过程中与轮询算法具有相同的控制信息量,不需要计算大量的状态信息。由于采用静态的仲裁器指针更新策略,完全避免了因指针同步引起的竞争冲突。仿真对比已有算法,APRR具有较优的网络性能。CICQ是一种能够很好地支持变长分组直接交换的结构。针对端口仲裁器在选择服务队列时会出现的VOQ队列吞吐率不公平的问题,提出了最小分组配额队列优先算法(MQF)。仿真证明,MQF算法在不同分组长度的业务到达下均具有良好的网络时延性能。最后,尝试在CICQ结构调度中引入成熟的传统IQ Crossbar的调度算法,实现混合结构调度,以达到较优的网络性能。这些分析对实际CICQ交换系统的设计和实施都有重要的参考价值和指导意义。

【Abstract】 The traffic on the Internet, especially the multimedia and the data traffic, is experiencing explosive growth which has led to a greater need for high-speed routers that have over 1-Tbit/s throughput. Switching Fabric, the kernel component in the router architecture, has a direct effect on the performance of routers. The major role of scheduling algorithm is to match the input ports to output ports, and resolve the resource contentions in the switching fabrics. Therefore scheduling algorithm has great influences on the performance of the switching fabrics. It is significant to make research on scheduling algorithm for large capacity switching fabrics.In this dissertation, we make our research on the Buffered Crossbar switch architecture. With today’s ASIC technology, a large number of memories can be easily implemented in a single chip. This makes the Buffered Crossbar switch a more attractive solution than the traditional input-queued Crossbar switch because of the simplicity of the Buffered Crossbar switch scheduling. The distributed character of this structure is suitable to be applied in large capacity router design.At first, the performance of the combined input-crosspoint-queued Crossbar (CICQ) switch is analyzed. Then a summary of existing research works has been made and by means of simulation, we analyzed the characters of some important factors such as crosspoint buffer size, speedup, switch size etc., which would affect the delay and throughput performance of CICQ architecture.Second, scheduling algorithm for CICQ is studied. On the basis of the summary of existing scheduling schemes, our research work is discussed in two different ways: the cells switch and the direct variable-length packets switch. In cell-mode, it can be found that pointers may be synchronous among arbiters in each side as well as the traditional Crossbar scheduler, which make the cells congestion at internal links. So we propose a fully asynchronous pointer round-robin scheduling scheme, called APRR scheme. In order to keep desynchronization, APRR makes the pointers at both the input and output sides totally different and updates them statically in clockwise and counter-clockwise at each time slot. Simulation results show that the throughput performance is significantly improved. As we know, CICQ architecture is suitable for switching variable-length packets directly. In this dissertation, in order to solve the problem about the throughput unbalance between traffics in every input port, which would appear due to using simple cell scheduling algorithm, a Minimum Quotient First -- MQF scheme is proposed. By choosing proper quotient, it takes advantage of the property of Buffered Crossbar that natively supports variable-length packets switch. Through simulation, MQF exhibits better performance than other existing schemes.At last, we tried to utilize the mature technique in traditional Crossbar to the CICQ switch architecture to achieve a high performance hybrid scheduling. This attempt uses complicated scheduling procedure to bargain a good performance. All of the analysis has been of great value to the design of CICQ switch.

  • 【分类号】TP393.01
  • 【被引频次】6
  • 【下载频次】371
节点文献中: