An evolution to crossbar switches with virtual output queuing and buffered cross points

Input queued (IQ) switch architectures with virtual output queues (VOQ) scale up to very high speeds and have been a subject of intense research in the past decade. VOQ IQ switches require switch matrix scheduling algorithms to match input ports to out ports. In this tutorial article, we present an overview of switch matrix scheduling for VOQ IQ switches with crossbar switch fabrics. We then describe what we believe will be the next generation of high-speed crossbar switches: the evolution of IQ switches to combined input and crossbar queued (CICQ) switches. With the continued increase in density of VLSI, sufficient buffering at crossbar cross points for one cell or packet has become feasible to implement. We show how CICQ switches have simple schedulers and result in lower delay than IQ switches. Both IQ and CICQ switches have unstable regions. We show how a threshold and bursting technique can feasibly achieve stability. We also show how CICQ switches are better suited (than IQ switches) for switching of variable-length packets such as IP packets. Many challenges remain in IQ and CICQ switches. In particular, the inclusion of QoS scheduling methods that are currently only suitable for output queued switches is a major open problem.     

高速信元交换调度算法研究

输入缓存交换结构的特点是缓存器和交换结构的运行速率与端口速率相等、实现容易,但存在队头阻塞(HOL),其吞吐率只有约58%.采用虚拟输出排队方法(VOQ)和适当的信元调度算法可消除HOL,使吞吐率达到100%.本文通过仿真对几种调度算法:PIM、iSLIP和LPF进行了全面地研究、比较和评价.     

保证100%吞吐率的两级组播交换结构

在路由器或交换机的交换结构中实现组播是提高组播应用速度的重要途径之一。传统的交叉开关结构(crossbar)组播调度方案有两种缺陷,一种是性能较低,另一种是实现的复杂度太高,无法满足高速交换的需要。该文提出了一个新的基于交叉开关的两级组播交换结构(TSMS),第1级是组播到单播的交换结构,第2级是联合输入和输出排队(CIOQ)交换,并为该结构设计了合适的最大扇出排队(FCN)优先-均匀分配中间缓存调度算法(LFCNF-UMBA)。理论分析和仿真实验都显示在该结构中,加速比低于22/(N+1)倍时吞吐率不可能实现100%;而采用LFCNF-UMBA调度算法,2倍加速比就可保证在任意允许(admissible)组播的吞吐率达到100%。     

Asymptotic Performance Limits of Switches With Buffered Crossbars Supporting Multicast Traffic

Input queued (IQ) switches exploiting buffered crossbars (CICQ switches) are widely considered very promising architectures that outperform IQ switches with bufferless switching fabrics both in terms of architectural scalability and performance. Indeed the problem of scheduling packets for transfer through the switching fabric is significantly simplified by the presence of internal buffers in the crossbar, which makes possible the adoption of efficient, simple and fully distributed scheduling algorithms. This paper studies the throughput performance of CICQ switches supporting multicast traffic, showing that, similarly to IQ architectures, also CICQ switches with arbitrarily large number of ports may suffer of significant throughput degradation under ldquopathologicalrdquo multicast traffic patterns. Despite the asymptotic nature of these results, the authors believe that they can contribute to a deeper understanding of the behavior of CICQ architectures supporting multicast traffic.     

一种可提供QoS保障的新型交换结构

本文采用带缓存交叉开关作为核心交换单元,构建了一种空分复用扩展的联合输入/交叉节点/输出排队(SDM-CICOQ)交换结构,从理论上证明了当扩展因子为2时,SDM-CICOQ交换结构可以获得100%的吞吐量,并且能够完全模拟输出排队(OQ)交换结构,从而能够提供服务质量(QoS)保障.本文还给出了一种层次化优先级调度(HPS)方案作为SDM-CICOQ交换结构调度机制的工程设计参考,仿真结果表明采用HPS调度方案SDM-CICOQ交换结构可获得良好的性能.     

一种公平输入排队调度算法

输入排队交换结构以其良好的可扩展性被越来越多的高速交换机和路由器所采用。当前的调度算法大都以牺牲公平性来换取最大的吞吐量。但随着对QoS支持的要求增强,适用于输入排队交换结构的高效、公平的调度算法成为迫切需要解决的问题。该文提出了一种具有公平性保证的基于虚服务量的公平调度算法。理论分析和计算机仿真都表明算法在信元时延和公平性方面都能提供较好的保证。算法还具有与iSLIP相同的较低通信开销,以及和iLQF相同的算法复杂度。因此,算法具有较好的实用性。     

Optimum scheduling and memory management in input queued switches with finite buffer space

The goal of this paper is to design optimal scheduling and memory management so as to minimize packet loss in input queued switches with finite input buffers. The contribution is to obtain closed-form optimal strategies that minimize packet loss in 2/spl times/2 switches with equal arrival rates for all streams. For arbitrary arrival rates, the contribution is to identify certain characteristics of the optimal strategy, and use these characteristics to design a near-optimal heuristic. A lower bound for the cost associated with packet loss for N/spl times/N switches is obtained. This lower bound is used to design a heuristic which attains near-minimum packet loss in N/spl times/N switches with arbitrary N. These policies reduce packet loss by about 25% as compared to the optimal strategy for the infinite buffer case. The framework and the policies proposed here apply to buffer-constrained wireless networks as well.     

RPA: a flexible scheduling algorithm for input buffered switches

This paper presents and evaluates a quasi-optimal scheduling algorithm for input buffered cell-based switches, named reservation with preemption and acknowledgment (RPA). RPA is based on reservation rounds where the switch input ports indicate their most urgent data transfer needs, possibly overwriting less urgent requests by other input ports, and an acknowledgment round to allow input ports to determine what data they can actually transfer toward the desired switch output port. RPA must be executed during every cell time to determine which cells can be transferred during the following cell time. RPA is shown to be as simple as the simplest proposals of input queuing scheduling, efficient in the sense that no admissible traffic pattern was found under which RPA shows throughput limitations, and flexible, allowing the support of packet-mode operations and different traffic classes with either strict priority discipline or bandwidth guarantee requirements. The effectiveness of RPA is assessed with detailed simulations in uniform as well as unbalanced traffic conditions and its performance is compared with output queuing switches and the optimal maximum weighted matching (MWM) algorithm for input-buffered switches. A bound on the performance difference between the heuristic weight matching adopted in RPA and MWM is analytically computed     

Cell Switching Versus Packet Switching in Input-Queued Switches

Input Queued (IQ) switches have been well studied in the past two decades by researchers. The main problem concerning IQ switches is scheduling the switching fabric in order to transfer packets from input ports to output ports. Scheduling is relatively easier when all packets are of the same size. However, in practice, packets are of variable length. In the current implementation of switches, variable length packets are segmented into fixed length packets—also knowns as cells—for the purpose of scheduling. However, such cell-based switching comes with some significant disadvantages: (a) loss of bandwidth due to the existence of incomplete cells; and (b) additional overhead of segmentation of packets and re-assembly of cells. This is a strong motivation to study packet-based scheduling, i.e., scheduling the transfer of packets without segmenting them. The problem of packet scheduling was first considered by Marsan They showed that under any admissible Bernoulli IID (independent and identically distributed) arrival traffic, a simple modification of the Maximum Weight Matching (MWM) algorithm achieves 100% throughput. In this paper, we first show that no work-conserving (i.e., maximal) packet-based algorithm is stable for arbitrary admissible arrival processes. Thus, the results of Marsan are strongly dependent on the arrival distribution. Next, we propose a new class of “waiting” algorithms. We show that the “waiting”-MWM algorithm is stable for any admissible traffic using the fluid limit technique. We would like to note that the algorithms presented in this paper are distribution independent or universal. The algorithms and proof methods of this paper may be useful in the context of other scheduling problems.     

一种基于Credit的变长分组并行交换网络调度算法

该文提出了一种新的并行分组交换(PPS)网络调度算法。该算法通过在解复用器处采用以变长分组为业务分配单元的方式消除了信元的乱序问题;通过采用Credit机制进行业务分配,实现了业务到各个交换平面完全公平的分配;各个并行交换单元采用组合输入输出排队,降低了对缓存和交换平面的加速要求,同时可以充分利用现有单Crossbar网络调度算法的研究成果。文中证明了该算法对业务分配的公平性,对高速缓存的需求量以及整个网络的稳定性,仿真进一步证明了该算法具有良好性能。     

NN based ATM cell scheduling with queue length-based priorityscheme

The asynchronous transfer mode (ATM) is the choice of transport mode for broadband integrated service digital networks (B-ISDNs). We propose a window-based contention resolution algorithm to achieve higher throughput for nonblocking switches in ATM environments. In a nonblocking switch with input queues, significant loss of throughput can occur due to head-of-line (HOL) blocking when first-in first-out (FIFO) queueing is employed. To resolve this problem, we employ bypass queueing and present a cell scheduling algorithm which maximizes the switch throughput. We also employ a queue length based priority scheme to reduce the cell delay variations and cell loss probabilities. With the employed priority scheme, the variance of cell delay is also significantly reduced under nonuniform traffic, resulting in lower cell loss rates (CLRs) at a given buffer size. As the cell scheduling controller, we propose a neural network (NN) model which uses a high degree of parallelism. Due to higher switch throughput achieved with our cell scheduling, the cell loss probabilities and the buffer sizes necessary to guarantee a given CLR become smaller than those of other approaches based on sequential input window scheduling or output queueing     

On the necessary and sufficient requirement of a CIOQ switch to emulate an Output Queued switch

There has been much interest to emulate the behavior of Output Queued switches. The early result of such attempts was reported by Prabhakar and McKeown using the CIOQ switches with speedup factor of 4. Subsequently, Stoica and Zhang and independently Chuang et al. showed that a speedup of 2 in conjunction with their scheduling schemes would be sufficient for CIOQ switches to emulate Output Queued switches.Additionally, Chuang et al. showed that in “Average Sense” a speedup of 2?1/N is necessary and sufficient for CIOQ to emulate Output Queued switch behavior.Our paper reports that in the “Strict Sense” a speedup of 2 is both necessary and sufficient. We show this requirement using examples for 2x2 and 3x3 switches. Then, with a constructed traffic pattern, it is proved that in the “Strict Sense” a speedup of 2 is necessary to emulate the behavior of an Output Queued switch for any switch size N.Combining this result with the previous scheduling schemes, we conclude that in the “Strict Sense”, a speedup of 2 is the necessary and sufficient condition to emulate the behavior of an Output Queued switch, using a CIOQ switch.Additionally, easing the assumptions and allowing the packet segmentation, it is shown that the speedup requirement to emulate the behavior of an Output Queued switch can be reduced to values even smaller than 2?1/N. For this case a lower bound of 3/2 and an upper bound of 2 is proved.     

Opportunistic Fair Scheduling in Wireless Networks: An Approximate Dynamic Programming Approach

We consider the problem of temporal fair scheduling of queued data transmissions in wireless heterogeneous networks. We deal with both the throughput maximization problem and the delay minimization problem. Taking fairness constraints and the data arrival queues into consideration, we formulate the transmission scheduling problem as a Markov decision process (MDP) with fairness constraints. We study two categories of fairness constraints, namely temporal fairness and utilitarian fairness. We consider two criteria: infinite horizon expected total discounted reward and expected average reward. Applying the dynamic programming approach, we derive and prove explicit optimality equations for the above constrained MDPs, and give corresponding optimal fair scheduling policies based on those equations. A practical stochastic-approximation-type algorithm is applied to calculate the control parameters online in the policies. Furthermore, we develop a novel approximation method—temporal fair rollout—to achieve a tractable computation. Numerical results show that the proposed scheme achieves significant performance improvement for both throughput maximization and delay minimization problems compared with other existing schemes.     

Packet-mode scheduling in input-queued cell-based switches

We consider input-queued switch architectures dealing at their interfaces with variable-size packets, but internally operating on fixed-size cells. Packets are segmented into cells at input ports, transferred through the switching fabric, and reassembled at output ports. Cell transfers are controlled by a scheduling algorithm, which operates in packet-mode: all cells belonging to the same packet are transferred from inputs to outputs without interruption. We prove that input-queued switches using packet-mode scheduling can achieve 100% throughput, and we show by simulation that, depending on the packet size distribution, packet-mode scheduling may provide advantages over cell-mode scheduling.     

输入排队结构交换机分组调度研究

以决定分组调度算法的交换结构为基础,从协调,减少和隔离输入排队交换结构中输入输出竞争裁决冲突的角度,分别讨论了VOQ,CIOQ,CICQ结构中的分组调度问题,并以当前最新的调度算法为例加以说明,进行了定性分析和定量对比,指出了具体有待研究的问题。随后讨论了最近才开始研究的光电混合结构中的分组调度问题。最后从交换结构和算法两个方面探讨了今后的研究方向和发展趋势。     

Matching output queueing with a combined input/output-queued switch

The Internet is facing two problems simultaneously: there is a need for a faster switching/routing infrastructure and a need to introduce guaranteed qualities-of-service (QoS). Each problem can be solved independently: switches and routers can be made faster by using input-queued crossbars instead of shared memory systems; QoS can be provided using weighted-fair queueing (WFQ)-based packet scheduling. Until now, however, the two solutions have been mutually exclusive-all of the work on WFQ-based scheduling algorithms has required that switches/routers use output-queueing or centralized shared memory. This paper demonstrates that a combined input/output-queueing (CIOQ) switch running twice as fast as an input-queued switch can provide precise emulation of a broad class of packet-scheduling algorithms, including WFQ and strict priorities. More precisely, we show that for an N×N switch, a “speedup” of 2-1/N is necessary, and a speedup of two is sufficient for this exact emulation. Perhaps most interestingly, this result holds for all traffic arrival patterns. On its own, the result is primarily a theoretical observation; it shows that it is possible to emulate purely OQ switches with CIOQ switches running at approximately twice the line rate. To make the result more practical, we introduce several scheduling algorithms that with a speedup of two can emulate an OQ switch. We focus our attention on the simplest of these algorithms, critical cells first (CCF), and consider its running time and implementation complexity. We conclude that additional techniques are required to make the scheduling algorithms implementable at a high speed and propose two specific strategies     

带偏射补偿机制的Birkhoff-von-Neumann交换机方案及其性能分析

Birkhoff-von-Neumann(BvN)交换机具有较低的执行复杂度和较高的吞吐量,但无法在业务突发的环境下提供性能保证。为此,提出一种带偏射的BvN(D-BvN)交换机制来增强交换机性能。D-BvN交换机通过平均业务矩阵的BvN分解,为每个虚电路(VC)提供均值带宽保证,同时通过偏射来处理业务突发。其主要思想是利用处于空闲状态的VC的闲置容量处理处于溢出状态的VC的溢出业务。具体地,偏射机制利用空闲VC的闲置容量完成两件事情:一是把溢出业务偏射到其他VC,二是给偏射业务提供到达目的端口的带宽。分析和仿真结果表明,所提方法不仅可以获得接近100%的输入负载吞吐量,而且具有较低的包乱序概率和较小的业务包延时。     

Approximating Fluid Schedules in Crossbar Packet-Switches and Banyan Networks

We consider a problem motivated by the desire to provide flexible, rate-based, quality of service guarantees for packets sent over input queued switches and switch networks. Our focus is solving a type of online traffic scheduling problem, whose input at each time step is a set of desired traffic rates through the switch network. These traffic rates in general cannot be exactly achieved since they assume arbitrarily small fractions of packets can be transmitted at each time step. The goal of the traffic scheduling problem is to closely approximate the given sequence of traffic rates by a sequence of transmissions in which only whole packets are sent. We prove worst-case bounds on the additional buffer use, which we call backlog, that results from using such an approximation. We first consider the NtimesN, input queued, crossbar switch. Our main result is an online packet-scheduling algorithm using no speedup that guarantees backlog at most (N+1)² /4 packets at each input port and each output port. Upper bounds on worst-case backlog have been proved for the case of constant fluid schedules, such as the N²-2N+2 bound of Chang, Chen, and Huang (INFOCOM, 2000). Our main result for the crossbar switch is the first, to our knowledge, to bound backlog in terms of switch size N for arbitrary, time-varying fluid schedules, without using speedup. Our main result for Banyan networks is an exact characterization of the speedup required to maintain bounded backlog, in terms of polytopes derived from the network topology     

Multicast scheduling for input-queued switches

We design a scheduler for an M×N input-queued multicast switch. It is assumed that: 1) each input maintains a single queue for arriving multicast cells and 2) only the cell at the head of line (HOL) can be observed and scheduled at one time. The scheduler needs to be: 1) work-conserving (no output port may be idle as long as there is an input cell destined to it) and 2) fair (which means that no input cell may be held at HOL for more than a fixed number of cell times). The aim is to find a work-conserving, fair policy that delivers maximum throughput and minimizes input queue latency, and yet is simple to implement. When a scheduling policy decides which cells to schedule, contention may require that it leave a residue of cells to be scheduled in the next cell time. The selection of where to place the residue uniquely defines the scheduling policy. Subject to a fairness constraint, we argue that a policy which always concentrates the residue on as few inputs as possible generally outperforms all other policies. We find that there is a tradeoff among concentration of residue (for high throughput), strictness of fairness (to prevent starvation), and implementational simplicity (for the design of high-speed switches). By mapping the general multicast switching problem onto a variation of the popular block-packing game Tetris, we are able to analyze various scheduling policies which possess these attributes in different proportions. We present a novel scheduling policy, called TATRA, which performs extremely well and is strict in fairness. We also present a simple weight-based algorithm, called WBA     

输入队列交换的公平可扩展调度系统

提出了一种基于输入队列交换的公平可扩展网络调度系统FSSA.通过将若干个容量较小的调度器合理连接并使其协同工作,构成多端口大容量网络交换调度系统,解决了单个调度器容量和端口数受集成电路工艺限制的问题.FSSA不仅速度高、规模可扩展而且易于硬件实现.环型连接、管线工作及公平调度技术的采用使FSSA在性能方面得到了进一步优化.仿真结果显示,FSSA的性能可与基于iSLIP、DSRR等算法的单片调度器相比拟,尤其在流量较大时,FSSA的性能明显优于单调度器性能.