An optical multi-mesh hypercube: a scalable optical interconnectionnetwork for massively parallel computing

A new interconnection network for massively parallel computing is introduced. This network is called an optical multi-mesh hypercube (OMMH) network. The OMMH integrates positive features of both hypercube (small diameter, high connectivity, symmetry, simple control and routing, fault tolerance, etc.) and mesh (constant node degree and scalability) topologies and at the same time circumvents their limitations (e.g., the lack of scalability of hypercubes, and the large diameter of meshes). The OMMH can maintain a constant node degree regardless of the increase in the network size. In addition, the flexibility of the OMMH network makes it well suited for optical implementations. This paper presents the OMMH topology, analyzes its architectural properties and potentials for massively parallel computing, and compares it to the hypercube. Moreover, it also presents a three-dimensional optical design methodology based on free-space optics. The proposed optical implementation has totally space-invariant connection patterns at every node, which enables the OMMH to be highly amenable to optical implementation using simple and efficient large space-bandwidth product space-invariant optical elements     

Optical multiple-access mesh-connected bus interconnects

We consider networks for interconnecting parallel computers that are parameterized by their degree (fan-out) of the switches that are used to build the network. Electronic networks tend towards the small fan-out/large diameter end of the spectrum due to limitations on the fan-out of electronic switches. The fan-out of systems of optical switches is anticipated to be significantly higher than electronic switches. Thus it has been suggested that optics can be used to build the complete network which is the most desirable of Clos's family of networks. Even optics, however, has limits on the fan-out of switches in any given configuration. Thus in this paper, we describe an implementation of a network with a significantly smaller degree and slightly larger network diameter than the complete network. The network that we implement is called the mesh-connected bus network. We remark, however, that it is essentially an implementation of the second most desirable network in the family of networks defined by Clos, which is widely referred to as the Clos network. We implement the network using novel free-space optical interconnection techniques along with a multiple-access scheme based on wavelength-division multiplexing     

Optical fiber interconnection for the scalable parallel computingsystem

In this paper, we discuss the optical fiber interconnection technologies applied in the two types of parallel processing systems: 1) a backplane interconnection in a parallel processor array system and 2) a computing cluster network. We have set up a parallel processor array system using optical fiber to make point-to-point interconnection between processor elements and are developing a low-cost virtual parallel optical fiber interconnection link (VPOFLink) complying with peripheral component interconnect (PCI) local bus specifications for the computing cluster. VPOFLink is integrated with the popular PCI bus interface in order to make the link hold the same bandwidth as that of the PCI bus. It was fabricated as an available peripheral device that can been inserted into the bus slots of commercial computers directly and can operate under the control of PCI bus. Also in this paper, we demonstrate the optical fiber link for a ring network and the architecture of the ring network     

Mapping a class of neural networks on k-ary n-cubes

In this paper we propose a scheme for mapping two important artificial neural network (ANN) models on the popular k-ary n-cube parallel architectures (KNCs). The scheme is based on generalizing the mapping of a bipartite graph onto the KNC architecture and thus can be adapted to any model whose computations can be represented by a bipartite task graph. Our approach is the first to adjust the granularity of parallelism so as to achieve the best possible performance based on properties of the computational model and the target architecture. We first introduce a methodology for optimal implementation of multi-layer feedforward artificial neural networks (FFANNs) trained with the backpropagation algorithm on KNCs. We prove that our mapping methodology is time-optimal and that it provides for maximum processor utilization regardless of the structure of the FFANN. We show that the same methodology can be utilized for efficient mapping of Radial Basis Function neural networks (RBFs) on KNCs. This revised version was published online in June 2006 with corrections to the Cover Date.     

TIONets: triangular interconnection of optical networks

We propose a multihop network called TIONet or triangular interconnection of optical networks. The graph is examined from routing and scaling approaches. A store and forward routing scheme is assumed     

一种可部署在重路由光网络中实现有效连接管理的策略——基于分布式代理与蚁群算法

在大型骨干光网络中, 建立一条从源节点到宿节点的路由至关重要。在集中式网络管理系统体系下, 网络中动态路由信息以及故障状态信息, 通过路由的“泛洪”机制, 被传送到其他节点上实现资源信息共享,这对于连接的迅速建立, 恢复路由的快速启动将是不利的。尝试采用一种启发式算法-蚁群策略, 快速提高光网络中故障恢复的速度。该路由算法平台采用了分布式蚁群算法自动修复故障点, 解决了重路由的快速收敛问题; 依靠分布式代理机制构建管理平面和数据平面相互通信的机制, 完成诸如链路信息素的更新等功能。设计了一个网络故障的场景, 通过实验数据表明, 采用蚁群的分布式路由策略, 可以更快地实现路由信息的汇聚, 找到最佳的恢复路由。     

A survey and comparison of wormhole routing techniques in a meshnetworks

The growing demand for high processing power in various scientific and engineering applications has made multiprocessing architectures increasingly popular. These multiprocessing systems consist of processing elements or nodes which are connected together by interconnection networks in various topologies. One of the design methodologies used for parallel machines has fed to the development of distributed memory message-passing concurrent computers, commonly known as multicomputers. They consist of many processing nodes that interact by sending messages (containing both data and synchronization information) over a communication link, between nodes. Thus, efficient communication in multicomputers is one of the important research areas in parallel computing today, and it depends on the underlying scheme for routing. For this reason it is essential to know which routing techniques are suitable and practical. Although an extremely wide number of routing algorithms have been proposed and implemented in hardware and software, it is difficult for the designer of a multicomputer to choose the best routing algorithm given a particular architectural configuration. In an attempt to overcome this difficulty, we present a survey and comparison of wormhole routing techniques in mesh interconnection networks. The mesh topology is important because of its scalability. Moreover, it has already been implemented in many commercial multicomputers     

Partitioned optical passive star (POPS) multiprocessorinterconnection networks with distributed control

This paper presents a partitioned optical passive star (POPS) interconnection topology and a control methodology that, together, provide the high throughput and low latency required for tightly coupled multiprocessor interconnection applications. The POPS topology has constant and symmetric optical coupler fanout and only one coupler between any two nodes of the network. Distributed control is based on the state sequence routing paradigm which multiplexes the network between a small set of control states and defines control operations to be transformations of those states. These networks have highly scalable characteristics for optical power budget, resource count, and message latency. Optical power is uniformly distributed and the size of the system is not directly limited by the power budget. Resource complexity grows as O(n) for the couplers, O(n√n) for transceivers, and O[√nlog(n)] for control. We present analysis and simulation studies which demonstrate the ability of a POPS network to support large scale parallel processing (1024 nodes) using current device and coupler technology     

Optoelectronic implementations of neural networks

The ability of optical systems to provide the massive interconnections between processors required in most neural network models, which constitutes their chief advantage for such applications, is discussed, focusing on holography. Because of the essential nonlinearity of the holographic connections, nonlinear processing elements are needed to perform complex computations. The use of GaAs hybrid optoelectronic processing elements is examined. GaAs is an excellent material for this purpose, since it can be used to fabricate both fast electronic circuits and optical sources and detectors. It is shown how a complete hybrid neural computer can be implemented using available technology developed for conventional computing. An experimentally demonstrated network in which optics plays an even larger role is described     

Adaptive alternate routing in WDM networks and its performance tradeoffs in the presence of wavelength converters

Routing in wavelength-routed all-optical WDM networks has received much attention in the past decade, for which fixed and dynamic routing methods have been proposed. Taking into account the observation that wavelength-routed all-optical WDM networks are similar to circuit-switched voice networks, except with regard to wavelength conversion, we propose an adaptive alternate routing (AAR) scheme for wavelength-routed all-optical WDM networks. A major benefit of AAR is that it can operate and adapt without requiring an exchange of network status, i.e., it is an information-less adaptive routing scheme. The scope of this work is to understand this scheme in its own right since no other dynamic routing schemes are known to have the information-less property. In this paper, we conduct a systematic study of AAR with regard to factors such as the number of converters, load conditions, traffic patterns, network topologies, and the number of alternate paths considered. We observe that the routing scheme with multiple alternate routes provides more gain at a lower load instead of requiring any nodes to be equipped with wavelength converters. On the other hand, the availability of wavelength converters at some nodes, along with adaptive routing, is beneficial at a moderate to high load without requiring all nodes to be equipped with wavelength converters. We also observed that a small number of alternate routes considered in a network without wavelength converters gives a much better performance than a network with full wavelength converters and fewer alternate routes. Throughout this study, we observed that the proposed adaptive alternate routing scheme adapts well to the network traffic condition.     

Optimal cluster-based routing scheme using node mobility in ad hoc networks

Ad hoc networks have a scalability problem. When the nodes of an ad hoc network increase in number or mobility, the amount of control traffic for routing increases and could cause traffic congestion. Cluster-based routing schemes have been proposed as a solution to this problem. Typical cluster-based ad hoc networks use a proactive routing scheme for intra-cluster routes and a reactive routing scheme for inter-cluster routes. In this study, we propose a new cluster-based routing scheme for ad hoc networks which makes use of the mobility of nodes. Nodes are divided into two groups on the basis of their mobility. For a route search within a cluster, a proactive routing scheme is used for low-mobility nodes and a flooding-based reactive routing scheme is used for high-mobility nodes. The required control traffic of the proposed scheme is analyzed and optimal parameters of the proposed scheme are derived from the analysis. The numerical results show that the proposed scheme produces far less control traffic than a typical cluster-based routing scheme.     

Modularity in neural computing

This paper considers neural computing models for information processing in terms of collections of subnetwork modules. Two approaches to generating such networks are studied. The first approach includes networks with functionally independent subnetworks, where each subnetwork is designed to have specific functions, communication, and adaptation characteristics. The second approach is based on algorithms that can actually generate network and subnetwork topologies, connections, and weights to satisfy specific constraints. Associated algorithms to attain these goals include evolutionary computation and self-organizing maps. We argue that this modular approach to neural computing is more in line with the neurophysiology of the vertebrate cerebral cortex, particularly with respect to sensation and perception. We also argue that this approach has the potential to aid in solutions to large-scale network computational problems - an identified weakness of simply defined artificial neural networks     

Proposal for st-Routing Protocol

A routing protocol chooses one of the several paths (routes) from a source node to a destination node in the computer network, to send a packet of information. In this paper, we propose a new routing protocol, which we call st-routing protocol, based on st-numbering of a graph. The protocol fits well in noisy environments where robustness of routing using alternative paths is a major issue. The proposed routing protocol provides a systematic way to retry alternative paths without generating any duplicate packets. The protocol works for only those networks that can be represented by biconnected graphs.     

Resource requirements and layouts for field programmableinterconnection chips

Field-programmable interconnection chips (FPIC's) provide the capability of realizing user programmable interconnection for any desired permutation. Such an interconnection is very much desired for supporting rapid prototyping of hardware systems and for providing programmable communication networks for parallel and distributed computing. An FPIC should realize any possible permutation of input to output pins via a set of programmable switches. In this paper, we show that any such architecture requires a minimum of Ω(n log n) switches, where Ω is the number of I/O pins. The result stems from an analysis of the underlying permutation network. In addition, for networks of bounded degree d, we prove an Ω(log_d-1 n) bound on the routing delay (maximum length of routing paths for specific I/O permutations) and an Ω(n log_d-1 n) bound on the average utilization of programmable switches used by the FPIC to implement a specific permutation. For the same type of networks, we prove an Ω(n log_d-1 n) bound on the number of nodes of the network. Furthermore, we design efficient architectures for FPIC's offering a wide variety of routing delays, high average programmable resource utilization, and O(n²)-area two-layer layouts. The proposed structures are called hybrid Benes-Crossbar (HBC) architectures and clearly exhibit a tradeoff between performance (routing delay utilization) and area of the layout     

用于机群系统的高速光互连网络接口卡设计

本文在已设计完成的用于计算机点到虚拟并行互连的光互连链路基础上,用现场可编程门阵列（ＦＰＧＡ）器件实现了数据的寻址和转发,设计完成了具有硬件器由功能的光互连环网的网络接口卡。该网络接口卡的旁路数据转发延迟为２１０ｎｓ,目的结点的数据接收处理延迟为３５０ｎｓ。     

平坦瑞利环境下多用户检测的量子神经网络方法

提出采用量子神经网络(QNN)方法在平坦瑞利环境下进行多用户检测的方法。量子神经网络是量子计算与人工神经网络(ANN)相结合的产物,由于利用量子并行计算和量子纠缠等特性从而克服了传统人工神经网络的固有缺点。研究结果表明:该算法具有较强的鲁棒性;能有效地抑制噪声干扰,克服远近效应,在平坦瑞利衰减下具有较好地误码性能。     

Hierarchical routing and traffic grooming in IP/MPLS-based ASON/GMPLS multi-domain networks

Routing, connection setup, and path computation are well-known problems in multi-domain networks, which have been largely analyzed in pure IP (packet) networks. In circuit-switched optical multi-domain networks, there remain, however, a number of routing and path computation challenges. Traffic grooming means combining a number of low-speed traffic streams so that the high capacity of each lightpath may be used as efficiently as possible, as path computation implements the core of the grooming function, it is obvious that solutions for the traffic grooming problem in optical multi-domain networks are still not sufficiently investigated. In this study we propose a methodology to address the problems of routing, connection setup, and traffic grooming in optical multi-domain networks, which adapts a two-level hierarchical routing scheme and full-mesh topology abstraction algorithm to improve routing scalability and lower inter-domain blocking probabilities; additionally our proposed methodology adapts a scheme for traffic grooming in DWDM multi-domain networks to improve the resources usage. To test our proposed methodology we propose a detailed IP/MPLS-based ASON/GMPLS multi-domain multilayer test framework.     

A resource-efficient and scalable wireless mesh routing protocol

By binding logic addresses to the network topology, routing can be carried out without going through route discovery. This eliminates the initial route discovery latency, saves storage space otherwise needed for routing table, and reduces the communication overhead and energy consumption. In this paper, an adaptive block addressing (ABA) scheme is first introduced for logic address assignment as well as network auto-configuration purpose. The scheme takes into account the actual network topology and thus is fully topology-adaptive. Then a distributed link state (DLS) scheme is further proposed and put on top of the block addressing scheme to improve the quality of routes, in terms of hop count or other routing cost metrics used, robustness, and load balancing. The network topology reflected in logic addresses is used as a guideline to tell towards which direction (rather than next hop) a packet should be relayed. The next hop is derived from each relaying node’s local link state table. The routing scheme, named as topology-guided DLS (TDLS) as a whole, scales well with regard to various performance metrics. The ability of TDLS to provide multiple paths also precludes the need for explicit route repair, which is the most complicated part in many wireless routing protocols. While this paper targets low rate wireless mesh personal area networks (LR-WMPANs), including wireless mesh sensor networks (WMSNs), the TDLS itself is a general scheme and can be applied to other non-mobile wireless mesh networks.     

Routing of 100 Gb/s words in a packet-switched optical networkingdemonstration (POND) node

This paper presents the design and experimental results of an optical packet-switching testbed capable of performing message routing with single wavelength time division multiplexed (TDM) packet bit rates as high as 100 Gb/s. The physical topology of the packet-switched optical networking demonstration (POND) node is based on an eight-node ShuffleNet architecture. The key enabling technologies required to implement the node such as ultrafast packet generation, high-speed packet demultiplexing, and efficient packet routing schemes are described in detail. The routing approach taken is a hybrid implementation in which the packet data is maintained purely in the optical domain from source to destination whereas control information is read from the packet header at each node and converted to the electrical domain for an efficient means of implementing routing control. The technologies developed for the interconnection network presented in this paper can be applied to larger metropolitan and wide area networks as well