一种动态分配虚拟输出队列结构的片上路由器

传统虚通道流控技术的片上路由器通过增加虚通道缓解排头阻塞引起的链路吞吐率下降以及网络拥塞的同时,面临缓冲区低利用率、仲裁开销较大等问题.而动态虚通道流控的片上路由器虽可通过动态管理缓冲单元,提高缓冲区利用率与链路吞吐率,但却不可避免流控与仲裁逻辑复杂度与开销的快速增长.为了提高链路吞吐率与缓冲区利用率,获得较好的性能与开销折中,提出一种动态分配虚拟输出队列结构的片上路由器DAVOQ,该结构通过快速链表动态组织虚拟输出队列,同时使用超前路由机制以简化仲裁逻辑,优化流水线.模拟与综合的结果表明,相比传统虚通道路由器,DAVOQ路由器改善报文传输延迟与吞吐率的同时,在0.13μm CMOS工艺下,节省了15.1%的标准单元面积与12.9%的漏电流功耗;而相比动态虚通道路由器,DAVOQ路由器能够以较小的吞吐率损失获得可观的延迟改善,同时节约15.6%的标准单元面积与20.5%的漏电流功耗.     

具有拥塞缓解策略的动态虚拟通道研究及其VLSI实现

虚拟通道技术改善了片上网络性能,却带来了巨大的面积与功耗开销.通过分析静态虚拟通道的不足,提出了基于拥塞缓解策略的动态虚拟通道结构.它采用链表方式组织缓冲,可以自动调整通道结构来适应各种流量负载:在较低流量下,该结构扩展通道队列深度,减小了报文传输延迟;在较高流量下,它增加虚拟通道数量,消除队列头阻塞与通道不足阻塞,并缓解拥塞现象发生,减少流反馈次数,提高了网络吞吐率.在90nm CMOS工艺下完成了DVC路由器的VLSI设计,与传统路由器相比,不仅改善了报文传输延迟与吞吐率,而且有效降低了面积与功耗开销.     

Torus网络中移动气泡流控及其自适应路由实现

在torus网络中气泡流控是一种有效、实用的死锁避免技术.关键气泡机制使用虚跨步技术,只需要使用一个报文缓冲区就可以避免torus网络中的环内死锁,但是可能存在阻塞.首先提出了伪报文协议,然后结合伪报文协议设计了移动气泡流控策略,克服了关键气泡不能移动时引起的阻塞.伪报文协议基于简单的请求-应答,移动气泡流控则使用传统的信用传输方法.采用该机制,路由器只需要最少两条虚通道,每条虚通道最少一个报文空间就可以实现无死锁完全自适应路由.通过对经典路由器进行适当修改,给出了实现移动气泡流控的方法.采用模拟器比较了各种气泡流控的性能,结果表明,移动气泡流控性能超出传统的气泡机制,而加入自适应机制后的性能明显高于其他非自适应方法,不仅降低了延迟,吞吐率也提高20%以上,最大幅度甚至达100%.     

低功耗片上网络路由器设计

针对片上网络路由器功耗问题,在系统级层次上对影响路由器功耗的虚通道数目、缓存深度和数据微片位数等关键因素进行了研究。提出了综合多种功耗关键因素以及虚拟通道共享交叉开关输入端口的功耗降低方法,设计实现了一种低能耗的NoC路由器。实验结果表明,与Alpha 21364路由器和IBM InfiniBand路由器相比,所设计的路由器具有较低的功耗。     

核心路由器中缓存容量问题的研究

目前核心路由器在网络拥塞时用以存储数据的缓存容量巨大,这虽然保持了极高的链路利用率,但使核心路由器的结构更为复杂。斯坦福法则认为路由器中缓存容量可减少为拇指法则的1%。该文指出斯坦福法则会引起较高的丢包率,提出在降低一定链路利用率的情况下,较小的缓存容量也可以获得较低的丢包率和较高的吞吐率。     

TCP流竞争拥塞及拥塞链路的缓存需求研究

针对流竞争拥塞,提出了一种拥塞分析模型FCCM(flow-competing congestion model),给出了TCP竞争流在拥塞链路上的分布特性,推导了流竞争拥塞发生的条件,进而分析了在流竞争拥塞发生时,路由器为维持拥塞链路100%利用率所需的最小缓存.分析结果表明,当流数目不确定时,应对流竞争拥塞所需的缓存将不大于流数目确定时经典BSCL(buffer sizing for congested interact links)方案中的最小缓存需求.     

DTN中依据报文质量的拥塞控制策略

为了解决受限网络环境所带来的低的递交率的问题,DTN(Delay Tolerant Networks)通常采用多拷贝路由(multiple-copy routing)机制来实现报文的有效递交。但在实际情况下,网络中节点的缓存是受限的,当网络中的总报文数大于所有节点的缓存容量和时,网络就发生了拥塞。针对网络拥塞,通过计算节点缓存中的报文质量(Quality of Message)以确定报文丢弃的优先级,合理地丢弃报文控制拥塞。仿真结果显示与其他拥塞控制策略相比,该路由策略能得到更好的网络性能。     

面向片上网络缓冲资源争用的路由器设计

针对片上网络典型路由器的缓冲资源利用率不高、大容量缓存设计受限等问题,在不增加缓存和虚通道的情况下,提出一种新的面向片上网络缓冲资源争用的路由器设计方案。在该路由器中,当某个输入端繁忙发生资源争用情况时,将阻塞数据包分配到其他拥有空闲缓存资源的输入端口,解决缓冲资源的争用问题,从而提高网络整体性能。SystemC仿真结果表明,相对于基本路由器,该路由器在热点模式和均衡模式下均具有较高的网络饱和率和吞吐量,尤其在热点模式下提高了约11.4%的饱和率。FPGA实现结果表明,该路由器的面积开销较小,能较好满足片上网络的应用需求。     

一种面向片上互连的自适应通道双缓冲延迟模型

随着集成电路工艺的等比例缩小,互连线延迟相对门延迟增加,导致报文在片上网络路由器之间的传输需要多个时钟周期。但是,在基于信用点流控策略中,物理链路中的寄存器在发生拥塞时不能够缓冲报文。因此,本文提出了一种自适应的通道双缓冲结构,能够在发生拥塞时缓冲报文。通过门级电路的设计和分析,根据逻辑努力方法建立了CDB的延迟模型。延迟模型的准确性利用Synopsys时序分析工具Prime Time在TSMC的65nm工艺库下被验证,两者相差不超过一个τ4。结果表明,在32nm工艺下,1mm长的半全局互连线通道双缓冲(CDB)和简单流水线(SPLS)所需要的级数相同。     

片上网络中面向报文有序传输的自适应路由算法研究

与确定性路由算法相比,自适应路由算法可以提高片上网络的通信性能,但是报文可能会无序到达。在目的节点对报文排序将会导致严重的面积和计算开销,甚至可能会抵消采用自适应路由算法带来的性能增益。为此,本文首先提出一种部分自适应路由算法,以满足报文的有序到达。然后,描述了对本文算法提供支持的路由器硬件结构。最后,在二维片上网络下对本文算法及确定性和自适应路由算法进行了性能评估和比较。与XY算法相比,本文算法显著降低了报文延时,提升了饱和点。同时讨论了对路由器面积和功耗影响。虽然路由器的功耗有所上升,但是由于报文交付性能提升,因此每个flit的能耗增长可忽略不计。     

SATATM的快速交换设计及业务量模型

SATATM利用卫星互连ATM网络,提供多点通信的信道。文中提出适应卫星信道特点的快速分组交换方法。交换网络内部采用总线式的虚拟直通以降低阻塞率,利用反馈控制和智能丢包提高信道利用率,以虚电路为单位控制输出缓冲以提高公平性。并且根据卫星信道的特点提出了SATATM的业务模型和控制方法。     

Quarter Load Threshold (QLT) flow control for wormhole switching in mesh-based Network-on-Chip

The fact that latency increases abruptly once the on-chip network is saturated indicates that it is necessary to devise an effective flow control strategy. Through tracing the status of the network buffer space we found that the payload of the on-chip network cannot get beyond an upper bound to avoid vicious congestion. Specifically, quarter of the total network buffer space is such a threshold, which is termed Quarter Load Threshold (QLT). Based on this fact we present the Quarter Load Threshold (QLT) flow control strategy. The performance of the proposed strategy is evaluated by the open source simulator Noxim [Noxim: Network-on-Chip Simulator, http://sourceforge.net/projects/noxim, 2008]. Simulation results show that the on-chip network runs smoothly and no serious congestion is encountered any more.     

Queueing network models of packet switching networks part 2: Networks with population size constraints

An overview of the application of product-form queueing networks to the performance analysis of store-and-forward packet communication networks is presented. Queueing network models with closed chains and other forms of population size constraints are considered. The modeling and analysis of networks with window flow controlled virtual channels are described first. Models for the performance analysis of strategies for nodal buffer management and permit-oriented network congestion control are then presented.     

Deterministic model and transient analysis of virtual circuits

A model for a virtual circuit in the form of a tandem of servers that process incoming packets using a FIFO (first-in, first-out) discipline is proposed. The service times are assumed to be known completely. These may differ from packet to packet and from server to server. The model permits a variety of buffer or transit time constraints to be incorporated into the model. Several results that help one to understand the transient behavior of a virtual circuit are presented. On the basis of these results, a number of schemes that may be used to determine the time when the next packet must be sent over the network are presented. Transit delay and throughput are used to evaluate a given schedule. Solutions are given for maximum throughput, minimum transit delay, and maximum throughput under transit delay constraints. It is expected that these results will have a substantial bearing on the study of congestion control policies in computer networks, particularly those based on predicting network behavior     

无线传感器网络的拥塞控制技术

无线传感器网络的多对一通信方式、无线链路的相互干扰、网络的动态变化和资源受限等特性,使得无线传感器网络容易出现拥塞,严重影响网络的QoS传输性能和生存周期,因此拥塞控制成为无线传感器网络服务质量保障机制的关键技术之一.在分析无线传感器网络特点的基础上,说明了拥塞检测和拥塞避免的策略,重点介绍和分析了基于速率控制、流量调度和传输调度等典型的拥塞解除算法,最后对拥塞控制技术的发展趋势进行了展望.     

一种提高TCP与UDP数据流公平性的拥塞控制机制

在UDP业务流逐步占据大部分网络带宽的情况下,如何在网络层对UDP包进行拥塞控制显得尤为重要。在网络拥塞时,由于UDP包本身缺乏反馈机制,将会产生严重的丢包或者UDP抢占TCP带宽的现象。文中提出了基于网络层的发送端缓冲队列管理的拥塞控制机制,可以均衡TCP和UDP的带宽使用,同时通过对路由器ICMP网络拥塞报文的处理,建立了高效的流量调节策略,并进行了网络仿真实验。实验结果表明,基于该机制的拥塞控制可以有效改善网络不同业务流带宽使用的不公平性。     

窄带Ad hoc网络端到端拥塞控制机理分析

基于窄带半双工信道的军用无线网络限制了ACK反馈信息的及时性和有效性,使得单程时延和交付概率等端到端传输指标成为发现网络拥塞的有效途径。利用M/G/1模型的排队方法分析和揭示了无线自组织网络中节点拥塞程度、链路层参数以及业务层面的性能参数之间的量化关系,从而为窄带无线自组织网络的拥塞控制协议设计提供了理论依据。     

A unified link-layer fault-tolerant architecture for network-based many-core embedded systems

Reliability is an important design concern for modern many-core embedded systems. Specifically, on-chip interconnecting systems are vulnerable to permanent channel faults and transient data transmission faults which may significantly impact the overall system performance. In this work, a Unified Link-layer Fault-tolerant NoC (ULF-NoC) architecture is proposed. ULF-NoC is developed for NoC equipped with bidirectional channels and features wormhole switching (instead of store-and-forward switching) and packet-based retransmission. An intelligent buffer controller is developed that does not require separate, dedicated buffer spaces to support packet retransmissions. Extensive simulations using both synthetic and real world data traffics demonstrated marked performance of the proposed ULF-NoC solution.     

A probabilistic approach for predictive congestion control in wireless sensor networks

Any node in a wireless sensor network is a resource constrained device in terms of memory, bandwidth, and energy, which leads to a large number of packet drops, low throughput, and significant waste of energy due to retransmission. This paper presents a new approach for predicting congestion using a probabilistic method and controlling congestion using new rate control methods. The probabilistic approach used for prediction of the occurrence of congestion in a node is developed using data traffic and buffer occupancy. The rate control method uses a back-off selection scheme and also rate allocation schemes, namely rate regulation （RRG） and split protocol （SP）, to improve throughput and reduce packet drop. A back-off interval selection scheme is introduced in combination with rate reduction （RR） and RRG. The back-off interval selection scheme considers channel state and collision-flee transmission to prevent congestion. Simulations were conducted and the results were compared with those of decentralized predictive congestion control （DPCC） and adaptive duty-cycle based congestion control （ADCC）. The results showed that the proposed method reduces congestion and improves performance.