Performance evaluation of the augmented data vortex switch fabric: An all-optical packet switched interconnection network

Modern high-performance computing systems require networks with high capacity, extremely high throughput and low latency in order to pass messages between thousands of processors and memory elements. Optical Interconnection Networks (OIN) offer a potentially viable solution to this requirement. An all-optical packet switched interconnection network called a Data Vortex (DV) switch has already been proposed by Yang et al. for the purpose of large scale photonic interconnections. For any interconnection network, fault tolerance and reliability are crucial issues, evaluation of which lacked attention for the case of the DV switch. In our earlier work we therefore presented the results for fault tolerance and reliability analysis of the primary DV switch. We also proposed a new Augmented Data Vortex (ADV) switch fabric, to improve the fault tolerance of the primary DV switch. The performance as regards fault tolerance of the ADV switch was computed and detailed results were obtained. In this paper, performance of ADV is investigated with reference to parameters such as latency and injection ratio (throughput) by means of numerical simulations. A uniform random traffic model has been used for the performance evaluation. The results obtained are compared with the results reported for the DV switch. The results show that the ADV switch with enhanced fault tolerance also improves the performance regarding latency. For same switch sizes (i.e. the same number of angles A, and height H) the injection ratios (throughput) for the DV and the ADV switches are comparable. Hence it can serve as a suitable candidate for high performance computing.     

The augmented data vortex switch fabric: An all-optical packet switched interconnection network with enhanced fault tolerance

Optical Multistage Interconnection Networks (OMINs) are capable of transmitting terabits of bandwidth per second, and they have been considered as possible solutions to the electronic communications bottleneck in interconnection networks. A novel architecture, the Data Vortex (DV) switch, has been proposed by Yang et al., as a scalable, ultra low latency, ultra high capacity, high throughput, low cross-talk and low BER, all-optical packet switching fabric that is a suitable candidate for use as an OMIN. For any interconnection network, its fault tolerance and reliability are crucial issues, which have lacked attention up to now in the case for a DV switch. In this paper we, therefore, present results of fault tolerance and reliability analysis of the primary DV switch, and propose (1) a new Augmented Data Vortex (ADV) switch fabric, to improve the fault tolerance of the primary DV switch. (2) The labelling and a numbering scheme, with detailed interconnections of nodes for the ADV switch is given. (3) A new self-routing procedure and a priority scheme for distributed control signalling in the ADV switch have been given. (4) For the first time, conversion of the 3-dimensional switch to an equivalent chained-MIN model, has been given, which is more suitable for later analysis of fault tolerance. (5) A multiplexing scheme at input ports and output ports which further enhances the fault tolerance of the ADV switch has been given. (6) Computation has been done of the reliability and fault tolerance of the new architecture via an analytical model. (7) Finally, comparison of the ADV switch architecture with the primary architecture (DV) in view of fault tolerance and reliability has been given, and hardware complexity and cost effectiveness have been studied.     

用于Peta级计算的低时延光互联技术综述

This paper reviews the recently developed optical interconnect technologies designed for scalable, low latency and high-throughput communications within datacenters or high performance computers. The three typical architectures including the broadcast-and-select based Optical Shared Memory Supercomputer Interconnect System (OSMOSIS ) switch, the defection routing based Data Vortex switch and the arrayed waveguide grating based Low-latency Interconnect Optical Network Switch (LIONS) switch are discussed in detail. In particular, we investigate the various loopback buffering technologies in LIONS and present a proof of principle testbed demonstration showing feasibility of LIONS architecture. Moreover, the performance of LIONS, Data Vortex and OSMOSIS with traditional state-of-the-art electrical switching network based on the Flattened-ButterFly (FBF) architecture in terms of throughput and latency are compared. The simulation based performance study shows that the latency of LIONS is almost independent of the number of input ports and does not saturate even at very high input load.     

Mesh-of-Trees and Alternative Interconnection Networks for Single-Chip Parallelism

In single-chip parallel processors, it is crucial to implement a high-throughput low-latency interconnection network to connect the on-chip components, especially the processing units and the memory units. In this paper, we propose a new mesh of trees (MoT) implementation of the interconnection network and evaluate it relative to metrics such as wire complexity, total register count, single switch delay, maximum throughput, tradeoffs between throughput and latency, and post-layout performance. We show that on-chip interconnection networks can provide higher bandwidth between processors and shared first-level cache than previously considered possible, facilitating greater scalability of memory architectures that require that. MoT is also compared, both analytically and experimentally, to some other traditional network topologies, such as hypercube, butterfly, fat trees and butterfly fat trees. When we evaluate a 64-terminal MoT network at 90-nm technology, concrete results show that MoT provides higher throughput and lower latency especially when the input traffic (or the on-chip parallelism) is high, at comparable area. A recurring problem in networking and communication is that of achieving good sustained throughput in contrast to just high theoretical peak performance that does not materialize for typical work loads. Our quantitative results demonstrate a clear advantage of the proposed MoT network in the context of single-chip parallel processing.     

A machine learning assisted optical multistage interconnection network: Performance analysis and hardware demonstration

Integration of the machine learning (ML) technique in all-optical networks can enhance the effectiveness of resource utilization, quality of service assurances, and scalability in optical networks. All-optical multistage interconnection networks (MINs) are implicitly designed to withstand the increasing high-volume traffic demands at data centers. However, the contention resolution mechanism in MINs becomes a bottleneck in handling such data traffic. In this paper, a select list of ML algorithms replaces the traditional electronic signal processing methods used to resolve contention in MIN. The suitability of these algorithms in improving the performance of the entire network is assessed in terms of injection rate, average latency, and latency distribution. Our findings showed that the ML module is recommended for improving the performance of the network. The improved performance and traffic grooming capabilities of the module are also validated by using a hardware testbed.     

New nodal design for high throughput data vortex optical interconnection network

This paper proposes a new three input nodal structure within the data vortex packet switched interconnection network. With additional optical switches, the modified architecture allows for two input packets in addition to a buffered packet to be processed simultaneously within a routing node. A much higher degree of parallel processing is allowed in comparison to previously proposed enhanced buffer node with two input processing or the original network node with single input processing. Unlike the previous contention prevention mechanism, the new network operates by introducing the packet blocking within the node if no exit path is available. This eliminates the traffic control signaling and the strict timing alignment associated with the routing paths which simplifies the overall network implementation. This study shows that both data throughput and the latency performance are improved significantly within the new network. The study compares the three input node with the two input node as well as the original single input data vortex node. Due to additional switch count and nodal cost, networks that support the same I/O ports and of the same cost are compared for a fair comparison. The limitation introduced by the blocking rate is also addressed. The study has shown that under reasonable traffic and network condition, the blocking rate can be kept very low without introducing complex controls and management for dropped packets. As previous architectures require operation under saturation point, the proposed architecture should also operate at reasonable level of network redundancy to avoid excessive packet drop. This study provides guidance and criteria on the proposed three input network design and operation for feasible applications. The proposed network provides an attractive alternative to the previous architectures for higher throughput and lower latency performance.     

Performances of the Data Vortex switch architecture under nonuniform and bursty traffic

The Data Vortex switch architecture has been proposed as a scalable low-latency interconnection fabric for optical packet switches. This self-routed hierarchical architecture employs synchronous timing and distributed traffic-control signaling to eliminate optical buffering and to reduce the required routing logic, greatly facilitating a photonic implementation. In previous work, we have shown the efficient scalability of the architecture under uniform and random traffic conditions while maintaining high throughput and low-latency performance. This paper reports on the performance of the Data Vortex architecture under nonuniform and bursty traffic conditions. The results show that the switch architecture performs well under modest nonuniform traffic, but an excessive degree of nonuniformity will severely limit the scalability. As long as a modest degree of asymmetry between the number of input and output ports is provided, the Data Vortex switch is shown to handle very bursty traffic with little performance degradation.     

Intelligent Query-Based Data Aggregation Model and Optimized Query Ordering for Efficient Wireless Sensor Network

Data aggregation algorithms play a primary role in WSN, as it collects and aggregates the data in an energy efficient manner so that the life expectancy of the network is extended. This paper intends to develop a query-based aggregation model for WSN using the advanced optimization algorithm called group search optimization (GSO). The proposed model is constructed in such a way that the querying order (QO) can be ranked based on latency and throughput. Accordingly, the main objective of the proposed GSO-based QO is to minimize the latency and maximize the throughput of WSN. The proposed data aggregation model facilitates the network administrator to understand the best queries so that the performance of the base station can be improved. After framing the model, it compares the performance of GSO-based QO with the traditional PSO-based QO, FF-based QO, GA-based QO, ABC-based QO and GSO-based QO in terms of idle time and throughput. Thus the data aggregation performance of proposed GSO-based QO is superior to the traditional algorithms by attaining high throughput and low latency.     

Priority packet switching in multistage interconnection networks

The throughput performance of a priority strategy for packet switching in multistage interconnection networks is studied. In this strategy, when packets at two or more inputs of a switching element contend for the same output line, the high priority packet has a higher probability of being selected for this output. By tuning these probabilities for different priority packets, a wide variety of throughput performances can be obtained.<>     

Modeling of Collision Avoidance Protocols in Single-Channel Multihop Wireless Networks

Although there has been considerable work on the performance evaluation of collision avoidance schemes, most analytical work is confined to single-hop ad hoc networks or networks with very few hidden terminals. We present the first analytical model to derive the saturation throughput of collision avoidance protocols in multi-hop ad hoc networks with nodes randomly placed according to a two-dimensional Poisson distribution. We show that the sender-initiated collision-avoidance scheme achieves much higher throughput than the ideal carrier sense multiple access scheme with a separate channel for acknowledgments. More importantly, we show that the collision-avoidance scheme can accommodate much fewer competing nodes within a region in a network infested with hidden terminals than in a fully-connected network, if reasonable throughput is to be maintained. Simulations of the IEEE 802.11 MAC protocol and one of its variants validate the predictions made in the analysis. It is also shown that the IEEE 802.11 MAC protocol cannot ensure collision-free transmission of data packets and thus throughput can degrade well below what is predicted by the analysis of a correct collision avoidance protocol. Based on these results, a number of improvements are proposed for the IEEE 802.11 MAC protocol.     

An early congestion feedback and rate adjustment schemes for many-to-one communication in cloud-based data center networks

Cloud data centers are playing an important role for providing many online services such as web search, cloud computing and back-end computations such as MapReduce and BigTable. In data center network, there are three basic requirements for the data center transport protocol such as high throughput, low latency and high burst tolerance. Unfortunately, conventional TCP protocols are unable to meet the requirements of data center transport protocol. One of the main practical issues of great importance is TCP Incast to occur many-to-one communication sessions in data centers, in which TCP experiences sharp degradation of throughput and higher delay. This important issue in data center networks has already attracted the researchers because of the development of cloud computing. Recently, few solutions have been proposed for improving the performance of TCP in data center networks. Among that, DCTCP is the most popular protocol in academic as well as industry areas due to its better performance in terms of throughput and latency. Although DCTCP provides significant performance improvements, there are still some defects in maintaining the queue length and throughput when the number of servers is too large. To address this problem, we propose a simple and efficient TCP protocol, namely NewDCTCP as an improvement of DCTCP in data center networks. NewDCTCP modified the congestion feedback and window adjusting schemes of DCTCP to mitigate the TCP Incast problem. Through detailed QualNet experiments, we show that NewDCTCP significantly outperforms DCTCP and TCP in terms of goodput and latency. The experimental results also demonstrate that NewDCTCP flows provide better link efficiency and fairness with respect to DCTCP.     

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     

2D-Mesh片上网络中通信密集点优化方法

针对2D-Mesh结构片上网络中通信密集点引发的网络拥塞问题,提出了一种分散通信密集点负载的方法,对网络互连结构进行局部调整,增加与大通信量模块相连的路由器数目,并设计了一种基于区域的XY-YX路由算法。仿真结果表明,该方法可以有效地降低通信延时,增大吞吐量。     

移动IPV6在改进数据包发送路径模型下性能分析

针对现有移动IPV6中数据包发送代价较高的问题,对现有移动IPV6协议性能分析模型中数据包发送路径进行了改进,将访问路由发送至移动节点的数据包由代价很大的隧道发送改进为直接发送,避免了隧道代价因子对数据包发送代价的影响。同时分析对比了移动IPv6协议MIPV6、FMIPV6、HMIPV6和PMIPV6在现有模型下和改进数据包发送路径模型下的数据包发送代价和切换时延、吞吐量。理论分析与实验仿真表明:现有移动IPv6协议在改进数据包发送路径模型下的数据包发送代价和切换时延比在现有模型下低,吞吐量有所提高。该改进数据包发送路径模型适合应用于数据包传输速率较高的场合。     

DIRSMIN: a fault-tolerant switch for B-ISDN applications usingdilated reduced-stage MIN

We have developed and analyzed a dilated high-performance fault-tolerant fast packet multistage interconnection network (MIN). This new switch, (d,d')-DIRSMIN, uses dilation to improve performance and fault-tolerance of a network. The links at the input and output stages of the dilated banyan-based MIN are rearranged to create multiple routes for each source-destination pair in the network, after removing the first stage in the network. These multiple paths are link- and node-disjoint. This new MIN can provide low packet-loss probability and high reliability with very little hardware overhead, compared to d-dilated banyan networks (BN). Fault tolerance at low latency is achieved by transmitting multiple copies of each input-packet simultaneously using different routes. A multiple-priority scheme allows alternate paths to be explored simultaneously, which results in higher throughput and reliability under both fault-free and faulty conditions. This guarantees that high throughput is maintained even in the presence of a fault. Throughput is analyzed using analytic and simulation methods; this new design has considerably higher performance in the presence of a permanent faulty switching-element (SE) or link, in comparison to dilated networks. Under non-faulty conditions, both analytic and simulation results show that a (d,d)-DIRSMIN performs better than the original dilated BN with the same SE complexity. We analyze the network reliability and show that the new design has superior reliability compared to competing proposals. In particular, this new design is considerably better than the SEN+, the best known thus far     

A multiple relay‐based medium access control protocol in multirate wireless ad hoc networks with multiple beam antennas

The advanced technique of multiple beam antennas is recently considered in wireless networks to improve the system throughput by increasing spatial reuse, reducing collisions, and avoiding co‐channel interference. The usage of multiple beam antennas is similar to the concept of Space Division Multiple Access (SDMA), while each beam can be treated as a data channel. Wireless networks can increase the total throughput and decrease the transmission latency if the physical layer of a mobile node can support multirate capability. Multirate wireless networks incurs the anomaly problem, because low data rate hosts may influence the original performance of high data rate hosts. In this work, each node fits out multiple beam antennas with multirate capability, and a node can either simultaneously transmit or receive multiple data on multiple beams. Observe that the transmitting or receiving operation does not happen at the same time. In this paper, we propose a multiple relay‐based medium access control (MAC) protocol to improve the throughput for low data rate hosts. Our MAC protocol exploits multiple relay nodes and helps the source and the destination to create more than one data channel to significantly reduce the transmission latency. Observe that low data rate links with long‐distance transmission latencies are distributed by multiple relay nodes, hence the anomaly problem can be significantly alleviated. In addition, the ACK synchronization problem is solved to avoid the condition that source nodes do not receive ACKs from destination nodes. An adjustment operation is presented to reduce unnecessary relay nodes during the fragment burst period. Finally, simulation results illustrate that our multiple relay‐based MAC protocol can achieve high throughput and low transmission latency. Copyright © 2009 John Wiley & Sons, Ltd.     

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。     

A Traffic Scheduling Technique for Metropolitan Area Gateways

In the future, metropolitan area networks (MAN's) will provide high bandwidth interconnection between local access networks in high density office, industrial, and residential settings. In these systems, the metropolitan area gateway (MAG) is responsible for coordinating the flow of traffic in and out of the metropolitan subnetwork. In certain environments, it is expected that the traffic will possess distinct characteristics. In this paper, a technique called SWIFT is proposed which efficiently schedules these traffic flows when the local access network is based upon broadband CATV-type cable technology. SWIFT satisfies the important constraint that it is compatible with existing CSMA/CD packet data network adapters which means that it may be implemented in previously installed systems. In this paper, a throughput model is given for the proposed technique. Also, an analytic model is given which approximates the mean remote delay when the traffic is moderately to heavily remote-bound. The performance results presented indicate that a system operating the SWIFT algorithm can achieve much reduced delays at higher throughput than would otherwise be possible.     

Handover in multihop cellular networks [Accepted from Open Call]

This article introduces various handover scenarios in multihop cellular networks. In addition, this article presents handover schemes where relay stations are located either inside a cell or on the boundary between two adjacent cells and investigates the effects of the deployment position of relay stations to handover performance. The simulation results show that multihop cellular networks for both deployment scenarios can achieve 90 percent throughput increase over single-hop cellular networks. The results also show that the overall throughput of the multihop cellular networks with relay stations inside a cell is higher than for those with relay stations on the boundary between two adjacent cells, whereas the opposite is observed for the throughput of cell-boundary users. The intercell handover latency in multihop cellular networks is increased by 20 ~ 56 percent compared with that in singlehop cellular networks because of the increased number of handovers and signaling overhead. However, by deploying relay stations on the boundary between two adjacent cells, the service-interruption time caused by inter-cell handover is reduced by 80 percent compared with that of single-hop cellular networks.     

一种有效的水声通信网络中用于半双工信道的ARQ协议

自动重传请求(ARQ)在保证通信网络可靠传输中起着关键的作用。但是，在水声通信网络中，由于水声信道的半双工以及信道的长传播时延特性，使得水声通信网络的吞叶性能受到极大的制约。本文针对水声信道长传播时延这一特点，利用扩频通信技术，将单个信道分成两个子信道；一个用于前向信道(从发送方到接收方)，另一个用于反向信道(从接收方到发送方)，从而使得对每个分组数据的响应时间大大缩短，提出了一种有效的适合于半双工特性的ARQ协议。理论分析和仿真结果表明，相对于传统的SW ARQ协议而言，该协议在水声信道下具有更优的吞吐性能。