All-optical Network Architecture Research Articles

Fast and scalable all-optical network architecture for distributed deep learning

With the ever-increasing size of training models and datasets, network communication has emerged as a major bottleneck in distributed deep learning training. To address this challenge, we propose an optical distributed deep learning (ODDL) architecture. ODDL utilizes a fast yet scalable all-optical network architecture to accelerate distributed training. One of the key features of the architecture is its flow-based transmit scheduling with fast reconfiguration. This allows ODDL to allocate dedicated optical paths for each traffic stream dynamically, resulting in low network latency and high network utilization. Additionally, ODDL provides physically isolated and tailored network resources for training tasks by reconfiguring the optical switch using LCoS-WSS technology. The ODDL topology also uses tunable transceivers to adapt to time-varying traffic patterns. To achieve accurate and fine-grained scheduling of optical circuits, we propose an efficient distributed control scheme that incurs minimal delay overhead. Our evaluation on real-world traces showcases ODDL’s remarkable performance. When implemented with 1024 nodes and 100 Gbps bandwidth, ODDL accelerates VGG19 training by 1.6× and 1.7× compared to conventional fat-tree electrical networks and photonic SiP-Ring architectures, respectively. We further build a four-node testbed, and our experiments show that ODDL can achieve comparable training time compared to that of an ideal electrical switching network.

RAMP: A flat nanosecond optical network and MPI operations for distributed deep learning systems

Distributed deep learning (DDL) systems strongly depend on network performance. Current electronic packet switched (EPS) network architectures and technologies suffer from variable diameter topologies, low-bisection bandwidth and over-subscription affecting completion time of communication and collective operations. We introduce a near-exascale, full-bisection bandwidth, all-to-all, single-hop, all-optical network architecture with nanosecond reconfiguration called RAMP, which supports large-scale distributed and parallel computing systems (12.8 Tbps per node for up to 65,536 nodes). For the first time, a custom RAMP-x MPI strategy and a network transcoder is proposed to run MPI collective operations across the optical circuit switched (OCS) network in a schedule-less and contention-less manner. RAMP achieves 7.6-171× speed-up in completion time across all MPI operations compared to realistic EPS and OCS counterparts. It can also deliver a 1.3-16× and 7.8-58× reduction in Megatron and DLRM training time respectively while offering 38-47× and 6.4-26.5× improvement in energy consumption and cost respectively.

Flat Ball: Dynamic topology for energy management of optical interconnection networks in data centers

Cloud computing and web applications have further highlighted the need for powerful data centers with large bandwidth and low power consumption. In order to provide such a large bandwidth, today's electronic data centers require high power levels. Furthermore, communications and devices are not always used to their full potential. Optical networks can be a proper option for data centers, as they can offer large and variable bandwidth with lower power consumption compared to their electronic counterparts. In this regard, a dynamic all-optical network architecture is proposed in this paper, with bandwidth reconfiguration capability and low latency to reduce energy consumption. This proposed architecture, called Flat Ball, offers dynamic bandwidth utilizing passive optical devices with low power consumption and latency. Under realistic data center traffic scenarios, the latency of this network is up to 50% lower than its electrical counterpart, while providing a much higher throughput and consuming up to 60% less power.

Prospects and research issues in multi-dimensional all optical networks

Research into all optical network (AON) technology has been ongoing over the past decade, and new features are constantly being developed. The advantages of AON include large-bandwidth provisioning, low-latency transmission and low energy consumption. The basic concept underlying AON is transmission of data signals entirely through the optical domain from source to destination nodes, with no optical-electrical-optical (O-E-O) conversion at intermediate nodes. The technologies used to implement AON have undergone a series of evolutions, which encompass time division multiplexing (TDM), frequency division multiplexing (FDM), and space division multiplexing (SDM). Multi-dimensional AON (MD-AON), which leads the trend of AONs future architecture, provides a vibrant state for emerging applications such as cloud computing and Internet of Things (IoT). In this article, we review the evolution of AON architectures based on the different all optical switching and multiplexing technologies (i.e., TDM, FDM, and SDM), which is one of the main areas of focus in this article. The other main area is detailed discussion of implementations such as data plane and control plane technologies as well as resource optimization technologies for realizing AON. We also introduce several AON testbeds with their compositions and functions, and some potential application scenarios that can be implemented based on these testbeds

Optimized Radio-over-Fiber-based cellular backhauling strategy using genetic algorithms and Pareto fronts

Radio-over-Fiber (RoF)-based cellular backhauling consitutes a promising solution to realize a cost-effective evolution of radio access networks towards higher radio-cells densification. In this paper, we recall the main characteristics of an innovative cellular backhauling all-optical network architecture called Generic Radio-over-Fiber Access Network (GeRoFAN) aiming to federate 4G system radio cells. However, the transport of multiple radio channels over an analog RoF link is subject to optical crosstalk and intermodulation distortions causing a decrease of the wireless system capacity. Supported by an analytical modeling of the relevant optical transmission limitations, we propose an impairment-aware radio cellular backhauling strategy for GeRoFAN able to maximize the radio cellular capacity while achieving an efficient use of optical resources. This optimization problem is solved through Pareto-based Genetic Optimization (PaGeO), a multi-objective meta-heuristic based on evolutionary algorithms and exploiting the concept of Pareto front. Numerical results indicate that PaGeO, applied to LTE cellular system, outperforms alternative backhauling strategies by achieving an excellent compromise between preserving the radio capacity of the system and the number of required optical channels. We also highlight how the optimal backhauling strategy calculated using PaGeO enables GeRoFAN to take full benefit from overlaying several radio channels per cell site.

Routing in Coloured Sparse Optical Tori by Using Balanced WDM and Network Sparseness

The authors present a WDM (Wavelength-Division Multiplexing) based all-optical network architecture, and study scheduled routing on it. Their architecture can be seen as a communication system of parallel multi-core computer or a large-scale high bandwidth routing switch of e.g., telecommunication network. The goal is to construct such a scalable architecture and a supporting routing protocol for it so that no electro-optical conversions are needed on the routing paths, all packets are routed along one of the shortest paths, processor nodes can inject packets constantly into the network, and all the packets injected into the routing machinery reach their targets without collisions. The authors’ CSOT is a sparse network. A large fraction of the nodes are intermediate nodes instead of processor nodes. Only the processor nodes are sources and sinks of packets. The number of all nodes is and is the number of processor nodes in our construction. For scheduled routing to work, the authors consider routing problems as a set of h-relations. They achieved work-optimal routing of -relations for a reasonable size of . The efficiency of routing is based on routing latency hiding which is made possible by WDM and sparseness based increase bandwidth per processor node.

All-optical cross-bar network architecture using TOAD based interferometric switch and designing of reconfigurable logic unit

The design of all-optical 2 × 2 Terahertz Optical Asymmetric Demultiplexer (TOAD) based interferometric switch is proposed and described in this manuscript. Numerical simulation has been done to achieve the performance of the switch. Using this 2 × 2 TOAD based switch, cross-bar network architecture is designed. A reconfigurable logic unit is also proposed in this manuscript, which can perform 16-Boolean logical operations.

Resilience of all-optical network architectures under in-band crosstalk attacks: a probabilistic graphical model approach

An important question for secure all-optical networks (AONs) is how to incorporate security against attacks in the design and engineering of network architectures. In this work, we study the resilience of AON architectures under in-band crosstalk attacks. Crosstalk attack propagation depends on both optical devices at the physical layer and wavelength usage at the network layer. This motivates us to employ probabilistic graphical models to model attack propagation. At the physical layer, we use a directed probabilistic graph (Bayesian Belief Network) to model the attack propagation under static network traffic and a given source of attack. At the network layer, we use an undirected probabilistic graph to represent the probability distribution of active connections in the network. The cross-layer model is obtained by combining the physical- and the network-layer models into a factor graph representation. Graphical models provide an explicit representation of interactions between the physical- and the network layer. Furthermore, graphical models facilitate derivations of analytical results on resilience with respect to physical-layer vulnerability, physical topology, and network load. Specifically, we derive bounds on the network resilience for regular topologies. For ring, star, and mesh-torus networks with link-shortest path routing and all-to-all traffic, we show that the average network resilience loss grows linearly with respect to the network load when the network load is small, and polynomially with respect to the probability of attack propagation from node to node along the attacker's route. In addition, numerical results suggest that the sum-product algorithm based on the factor graph representation can be used for computationally efficient evaluation of network resilience for irregular/large topologies

ARTEMIS: 40-gb/s all-optical self-routing node and network architecture employing asynchronous bit and packet-level optical signal processing

A 40-Gb/s asynchronous self-routing network and node architecture that exploits bit and packet level optical signal processing to perform synchronization, forwarding, and switching in the optical domain is presented. Optical packets are self-routed on a hop-by-hop basis through the network by using stacked optical tags, each representing a specific optical node. Each tag contains necessary control signals for configuring the node-switching matrix and forwarding each packet to the appropriate outgoing link and onto the next hop. In order to investigate the feasibility of their approach physical-layer simulations are performed, modeling each optical subsystem of the node showing acceptable signal quality and end-to-end bit error rates. In the All-optical self-RouTer EMploying bIt and packet-level procesSing (ARTEMIS) control plane, a timed/delayed resource reservation-based signaling scheme is employed combined with a load-balancing feedback-based contention-avoidance mechanism that can guarantee a high performance in terms of blocking probability and end-to-end delay

QoS Differentiation in Asynchronous Bufferless Optical Packet Switched Networks

Optical Packet/Burst Switched networks (OPS/OBS) have emerged as promising all-optical network architectures for future core networks due to good resource utilization and the ability to handle high line speeds in a cost-efficient way. Regarding these networks, Quality of Service (QoS) differentiation should be provided in order to give sufficient QoS to real-time applications, and to utilize network resources optimally. In this paper, we present three fundamental different QoS differentiation schemes suitable for asynchronous bufferless OPS. We present analytical models and derive explicit results for the packet loss rates. We then present a general evaluation framework followed by a comparison study of the considered QoS differentiation schemes. Simulation and analytical results show that QoS differentiation schemes based on preemption have better performance than drop based and access-restriction based QoS differentiation schemes.

Broadcasting, Scalability, and Reconfigurability Aspects in an All-Optical Network Architecture

All-optical networks are a sound answer to the huge data traffic demand expected for the forthcoming information society. Interesting networking aspects are the broadcasting, scalability, and reconfiguration in an all-optical environment. These issues are addressed in the present article for an all-optical network called MATRIX, which is time slotted and uses packet switching. The architecture enables the all-optical interconnection of a large number of nodes with both a small number of wavelengths and wavelength continuity based on a consequent exploitation of wavelength-division multiplexing and space-division multiplexing (i.e., multiple fibers per cable are used). For an efficient use of the resources, each node is equipped with multiple transmitters and receivers, respectively. Two different broadcasting schemes are investigated. An analytical model is developed, and the corresponding simulation results are in good agreement. It is further shown how different networks of arbitrary size can be interconnected to larger all-optical network clusters, thereby providing scalability not only with respect to the number of nodes but also to the geographical extension. Finally, wavelength reassignments allow change of optical paths through the network in case of changing traffic patterns or network failures in order to best exploit the available network resources. The total number of possible configurations is determined by the theory of Latin Squares.

Security issues in all-optical networks

All-optical networks are emerging as a promising technology for terabit per second class communications. However, they are intrinsically different from electro-optical networks, particularly because they do not regenerate signals in the network. The characteristics of all-optical network components and architectures manifest new and still unstudied security vulnerabilities but also offer a new array of possible countermeasures. We focus on two types of attacks on the physical security of the network: service disruption, which prevents communication or degrades quality of service (QoS), and tapping, which compromises privacy by providing unauthorized users access to data which may be used for eavesdropping or traffic analysis.

All-optical packet-switched metropolitan-area network proposal

A proposal for developing a precompetitive, commercially viable all-optical network architecture using dense optical wavelength division is described. The objective was to demonstrate a packet-switched high-capacity all-optical LAN/MAN network based on wavelength-division multiplexing (WDM) and capable of supporting several hundred to a thousand nodes, each requiring gigabit-per-second throughputs. The discussion covers the rationale and objectives of the program; tunable components, optical amplifiers, network protocol, and application development; network interconnection studies; and testbed construction.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>