Dependable optical multistage interconnection networks

多级互连网络在多处理器系统中用于将处理器与其他处理器或内存模块连接起来。这些网络在低延迟和高成本的网络（如crossbar）和高延迟和低成本的网络（如共享总线）之间提供了一种折衷方案。此外，可以对MINs进行流水线化，以提供与特定流量模式的交叉条相当的带宽。具有电光开关元件（se）的最小值比电子最小值提供更高的信道带宽和更低的通信延迟。在光多级互连网络OMINs的结构、运行和性能方面已经进行了大量的工作。但是，在OMIN中共享一个交换机的两条路径可能会在交换机内的一条路径到另一条路径之间出现一些不希望出现的耦合。这就是所谓的光串扰。由于串扰的存在，传统的由电子交换实现的路由调度算法不适用于OMIN。在本研究中，我们将考虑如何有效地通过OMIN调度和路由消息。我们也会尝试提出比以前提出的设计更好的新的OMIN设计。我们将把我们的调查扩展到OMINs中的其他通信模式，如多播、广播、所有对所有的个性化交换和八卦。我们还将研究OMIN的容错方面。即尝试研究具有内置冗余以容忍交换元件故障和处理器故障的不同OMIN设计，并提出能够容忍故障的可靠通信算法和可靠OMIN。

Multistage Interconnection Networks (MINs) are used in multiprocessor systems to connect processors with other processors or memory modules.  These networks provide a compromise between networks of low latency and high cost such as the crossbar, and networks of high latency and low cost such as the shared bus.  Moreover, MINs can be pipelined to provide a bandwidth that is comparable to that of the crossbar for specific traffic patterns.  A MIN with electro-optic switch elements (SEs) offers a higher channel bandwidth and lower communication latency than the electronic MIN.  A large body of work has been conducted on the structure, operation, and performance of Optical multistage interconnection networks OMINs.  However, two paths, sharing a switch in an OMIN can experience some undesirable coupling from one path to another within the switch. This is called optical crosstalk.  Due to crosstalk, traditional routing and scheduling algorithms that are used for MIN implemented by electronic switching are not applicable for OMIN.       In this research, we will consider how to schedule and route messages through an OMIN efficiently. We would also try to come up with new OMIN designs that are better than previously proposed designs. We will extend our investigation to other communication patterns in OMINs such as multicasting, broadcasting, all-to-all personalized exchange, and gossiping.  We will also look at the fault-tolerant aspect of OMIN.  That is to try to investigate different OMIN designs that have built-in redundancy to tolerate switching element faults and processor faults, and to come up with dependable communication algorithms and dependable OMIN that can tolerate faults.