【作者】 刘焕淋；

【导师】 潘英俊；

【作者基本信息】 重庆大学 ， 光学工程， 2008， 博士

【摘要】 随着因特网的广泛应用和宽带WDM技术的实现,网络的瓶颈正从传输层转移到了网络层,尤其是在线路传输速度超过10Gb/s时,光交换技术被认为是减少网络节点处光/电/光(O/E/O)转换时间的一种非常有效的办法。为了能够完成全光交换,必须从出发点到目的地全部实现在光域的分组传输与交换。异步光分组交换网络是以光分组作为最小的交换颗粒,带宽可以按照用户的需求在分组级上灵活地分配,从而可以显著提高带宽的利用率,并且有效支持多粒度的数据业务,是能够满足不同业务需求的新一代全光交换网络。但是在实现方面,还有许多技术难题有待克服,主要包括异步光分组冲突解决方案、光节点的调度策略和算法、光分组头的处理技术和光分组的组装机制。其他关键技术还包括:各类全光逻辑器件实现、光分组路由算法、光分组编码方案、通道保护和恢复机制等。基于以上的分析,论文围绕上述异步光网络中的若干关键技术,主要对光分组的业务突发特性、光分组冲突解决方案、光节点的信道调度策略和光分组头的光域提取技术等方面进行了研究,主要研究内容如下:1)通过分析IP业务的突发性、自相似及对异步OPS网络业务模型描述和理论分析的影响,论文第二章进而研究了三种典型的分组长度分布特性对异步OPS性能影响,得到边缘节点应进行业务流量整形,使核心节点的分组突发性较小,以获得最佳的分组丢失率(PLR)性能。2)分析了常用FDL反馈型缓存解决方案特点,结合异步OPS冲突性较大的特点,研究了FDL缓存虚占用与交换式缓存的概念,并推导了FDL共享反馈缓存配置系统丢包性能近似表达式和波长变换与FDL共享反馈缓存联合配置系统丢包性能近似表达式。并研究了一种以FDL作为主要缓存,电子RAM作为辅助缓存的冲突解决配置方案和光电混合的FF-VF信道调度算法。3)由于IP业务网中存在大量对传输时延有严格要求的TCP和UDP数据,分析了基于VOQ型OPS交换节点的性能,研究了到达交换节点输入端口的长包和短包的区分方法和实现流程,设计了短包、长包队列的OPS节点结构,提出了应用于异步OPS节点的一种简单、灵活有效、时间复杂度较低的短包抢先调度PSPF算法。该算法调度器优先调度短包,短包请求调度的优先级最高,即使在长包的传输时间中,短包也能抢占长包的传输时间。4)分析了分组在FDL中缓存排序的概念,针对排序型分组可避免输出时不同FDL之间的内部竞争问题,增加延迟时限的到来,分组调度成功的概率较大的特点,研究了三种分组排序方案。ASPF是最短分组优先排序占用延时最短的FDL方案,分析了该方案复杂度较高;WSPF是较短分组优先排序占用延时最短的FDL方案,由于方案仅对一根输入光纤线的W个波长分组进行长度比较,算法复杂度有所减小;第三种方案是PLPF方案,较长的分组优先占用延时最小的FDL,该方案的复杂度也较低。分组的排序缓存使OPS性能提高,但系统的复杂性也增大。研究结果表明10%的排序率和分组100%的排序率获得的PLR性能相当。因此,用少量的短分组优先缓存排序,其它分组按先到先服务方案管理和使用有限数目的FDL缓存时,就能获得较为理想的PLR性能。5)分析了半导体放大器和延迟干涉仪特点,研究了基于SOA和改进可微调的D-I干涉仪的全光分组头提取方案,通过合理设置信头比特间隔、净荷比特间隔、SOA载流子恢复时间和信头与净荷的保护时间参数,可以实现在光域对比度达到15dB的光分组头提取方案,并分析和研究了提高信头提取对比度的一组优化参数值。该方案还具有结构简单,可扩展性好和易于集成的特点。更多还原

【Abstract】 As the widespread use of the Internet and broadband WDM technology realization, the network bottleneck is shifting from transporting layer to switching layer. Especially assuming that transmission line rates will continue to increase beyond 10Gb/s, optical switch techniques are expected to assist in releasing the network from undesirable latencies related to O/E/O conversions at the switching nodes. In order to switch the all-optical packet, transmit packets from source to destination all-optically, it is crucial to transmit and switch the packets in the all-optical domain from source to destination.Optical packet is the smallest switching granularity for asynchronous optical packet switching (AOPS) network. AOPS is the new generation of all-optical switching network fulfilling the diverse service application, which bandwidth is assigned agilely according to the user’s actual need with high bandwidth utility and supporting multi-granularities data service. But there have many technical difficult problem needing to solution for realizing the OPS technology, such as optical packet contention resolution, the scheduling strategy and algorithm for the AOPS node, optical header processing, optical packet assembling mechanism, etc. The other key technology for realizing AOPS include the all-optical logical devices realization, optical packet routing algorithm, optical packet coding scheme, optical channel protection and resumption mechanism, etc.Based on the above analysis, this dissertation mainly covers the following topics’research: optical packet service bursty features; optical packet contention resolution; optical channel scheduling strategy; optical header extraction in the optical domain.1) Considering the IP traffic is bursty and self-similar, we analyzied its performance influence establishing service description model and theoretic analysis mode for asynchronous OPS network. Then, how the three typical optical packets length distribution influence on the OPS performance is studied in the chapter 2 also. The results show that the OPS edge node need to reshape the service flow to make the service with low bursty coefficient entering the OPS core node and to achieve the optimal packet loss ratio (PLR) performance.2) The features of conventional FDL feedback buffering resolution are analyzed in the chapter 3. Considering the probability of asynchronous OPS packet conflicting is larger, the concept of FDL buffering virtual occupation and switch buffer are studied. The approximate expressions for OPS PLR performance are derivated under shared feedback FDL buffering configuration and wavelength converter with shared feedback FDL buffering configuration respectively. The contention resolution which FDL is used as primary buffer and electronical RAM is used as supplementary buffer and photoelectric hybrid First-Fit-Void-Filling (FF-VF) channel scheduling algorithm are put forward for the first time in chapeter 3.3) There have much TCP and UDP data having the strict transmission delay restriction in the IP service network. The OPS node performance with input queue worked on the virtual output queue is studied in the chapter 4. The realization way and flow of difference long length packet and short length packet is proposed in this chapter. The OPS node structure with long length packet and short length packet queue respectively is designed. A simple, effective, flexible and with low time complexity algorithm, namely preemptive short packet first (PSPF) scheduling is proposed for the OPS node. The algorithm stipulates the short packet have the hightest scheduling priority and the scheduler scheduling the short packet first through fair and foul, even if the long length packets are transmitting, the short length packets can preempt the long packet transmission time.4) The concept of packet FDL sorting is researched in the chapter 5. Due to the packet buffer sorting resolving the FDL internal contention and increasing the delay restriction, the probability of packets scheduling success is large. So, three packet buffer sorting schemes are studied. All-sorting shortest packet first (ASPF) scheme let the shortest packet occupy the shortest FDL with higest calculation complexity; Weak-sorted shorter packet first (WSPF) scheme let the shorter packet occupy the shortest FDL with the lesser complexity by only compareing W wavelength channel delay for one input fiber. The third packet buffer sorting scheme is partial-sorted longer packet first (PLPF) which let longer packet occupy first the shortest FDL. The packet buffer sorting imporves the OPS performance at the cost of system calculation complexity. The study result indicates that the performance of 10 percent packets buffer sorting correspond to the performance of 100 percent packets buffer sorting. Therefore, the scheme, which utilizes a small quantitiy of short packet buffer sorting and other packets with first-come-first-serve buffer management using limited number of FDL, can achieve perfect PLR performance.5) The feature of SOA and delay interferometer is analzied in the chapter 6. A novel all-optical packet header extraction scheme and structure based on SOA and improved tiny adjustable D-I interferometer is studied. By setting parameter suitably, such as header bit time interval, payload bit time interval, SOA carrier life time, guarantee time between header and payload, the scheme can achieve more than 15dB contrast ratio for the optical packet header extraction when the rates of header and payload are 2.5Gb/s and 40Gb/s respectively. The parameters of improved SOA and delay interferometer are analyzed and studied to optimize the performance of optical packet header extraction with the proposed scheme. In addition, the system structure is simple, highly scalable and photonic integration.更多还原