Revisiting flow-based load balancing: Stateless path selection in data center networks

Hash-based load-balancing techniques are widely used to distribute the load over multiple forwarding paths and preserve the packet sequence of transport-level flows. Forcing a long-lived, i.e., elephant, flow to follow a specific path in the network is a desired mechanism in data center networks to avoid crossing hot spots. This limits the formation of bottlenecks and so improves the network use. Unfortunately, current per-flow load-balancing methods do not allow sources to deterministically force a specific path for a flow.In this paper, we propose a deterministic approach enabling end hosts to steer their flows over any desired load-balanced path without relying on any packet header extension. By using an invertible mechanism instead of solely relying on a hash function in routers, our method allows to easily select the packet’s header field values in order to force the selection of a given load-balanced path without storing any state in routers.We perform various simulations and experiments to evaluate the performance and prove the feasibility of our method using a Linux kernel implementation. Furthermore, we demonstrate with simulations and lab experiments how MultiPath TCP can benefit from the combination of our solution with a flow scheduling system that efficiently distributes elephant flows in large data center networks.     

基于概要数据结构的全网络持续流检测方法

持续流是隐蔽的网络攻击过程中显现的一种重要特征，它不产生大量流量且在较长周期内有规律地发生，给传统的检测方法带来极大挑战。针对网络攻击的隐蔽性、单监测点的重负荷和信息有限的问题，提出全网络持续流检测方法。首先，设计一种概要数据结构，并将其部署在每个监测点；其次，当网络流到达监测点时，提取流的概要信息并更新概要数据结构的一位；然后，在测量周期结束时，主监测点将来自其他监测点的概要信息进行综合；最后，提出流持续性的近似估计，通过一些简单计算为每个流构建一个位向量，利用概率统计方法估计流持续性，使用修正后的持续性估计检测持续流。通过真实的网络流量进行实验，结果表明，与长持续时间流检测算法（TLF）相比，所提方法的准确性提高了50%，误报率和漏报率分别降低了22%和20%，说明全网络持续流检测方法能够有效监测高速网络流量。     

并行分组交换中基于分路的分布式负载平衡技术

负载平衡是并行分组交换的一项关键技术,目前已知在高速率查表后应用负载平衡技术可取得良好效果,但在OC768（40Gbps）甚至OC3072（160Gps）下进行包处理工作是很困难的．基于分路包处理和分路负载平衡的思想,提出一种基于分路的分布式负载平衡控制结构和DDPA算法,依据工程设计考虑,还提出基于丢包策略的DDPA算法,对它们的有效性做出证明并定量地计算出其关键参数．     

Differentiated Bandwidth Allocation with TCP Protection in Core Routers

Differentiated Services (DiffServ) networks categorize routers into edge routers and core routers. In core routers, one of the technological challenges is how to implement differentiated bandwidth allocation and TCP protection together with low complexity. We present an Active Queue Management (AQM) scheme called CHOKeW. A method is borrowed from a previous scheme, CHOKe, which draws a packet at random from the buffer, compares it with the arriving packet, and drops both if they are from the same flow. CHOKeW enhances the drawing function by adjusting the maximum number of draws based on the priority of the new arrival and the current status of network congestion. With respect to the number of flows, both the memory-requirement complexity and the per-packet-processing complexity for CHOKeW is O(1). An analytical model and multiple simulations are used to explain and evaluate CHOKeW. We show that CHOKeW is able to 1) support differentiated bandwidth allocation; 2) provide the flows in the same priority with better fairness than other conventional stateless AQM schemes such as RED and BLUE; 3) maintain high link utilization as well as short queue length; and 4) protect TCP flows by restricting the bandwidth share of high-speed unresponsive flows.     

Ordered lookup with bypass matching for scalable per-flow classification in layer 4 routers

In order to provide different service treatments to individual or aggregated flows, layer 4 routers in Integrated Services networks need to classify packets into different queues. The classification module of layer 4 routers must be fast enough to support gigabit links at a rate of millions of packets per second. In this work, we present a new software method OLBM to lookup multiple fields of a packet, in a dynamically pre-defined order, against the classification database. This algorithm also uses a technique called bypass matching and can classify packets at a rate of well over one million packets per second while scaling to support more than 300k flows. Complexity analysis and experiment measurements are also presented in this study.     

可信可控网络中跨自治域的流合作测量模型

当前网络环境下提出的流测量算法缺少必要的合作,存在测量覆盖度不高等问题。为此,提出一种新的基于全网合作的流测量模型,在自治域(AS)间通过标识已测量流和共享测量结果的方式,实现AS间的流合作测量,在AS内利用控制节点为路由器统一分配测量任务的方法,实现AS内最大化流测量覆盖度的目标。实验结果表明,该模型能够有效利用网络资源,平衡各个路由器的测量负载,提高测量覆盖度,具有比以往方法更好的性能。     

基于改进蚁群算法的网络负载均衡路由优化

最短路径优先的路由控制策略在解决突发业务模式下的拥塞问题上存在不足,针对该问题,采用随机选择尺度因子实现负载均衡,增加路由器向所有相邻路由器转发分组的机会。设计路由算法模拟器,完成改进蚁群网络路由优化算法在4种典型网络上的仿真。以CHINANET为例的对比实验表明,在重负载情况下,改进的蚁群网络路由优化算法能获得较高的吞吐量、较低的平均时延与丢包率。     

FlowTrace: measuring round-trip time and tracing path in software-defined networking with low communication overhead

In today’s networks, load balancing and priority queues in switches are used to support various quality-of-service (QoS) features and provide preferential treatment to certain types of traffic. Traditionally, network operators use ‘traceroute’ and ‘ping’ to troubleshoot load balancing and QoS problems. However, these tools are not supported by the common OpenFlow-based switches in software-defined networking (SDN). In addition, traceroute and ping have potential problems. Because load balancing mechanisms balance flows to different paths, it is impossible for these tools to send a single type of probe packet to find the forwarding paths of flows and measure latencies. Therefore, tracing flows’ real forwarding paths is needed before measuring their latencies, and path tracing and latency measurement should be jointly considered. To this end, FlowTrace is proposed to find arbitrary flow paths and measure flow latencies in OpenFlow networks. FlowTrace collects all flow entries and calculates flow paths according to the collected flow entries. However, polling flow entries from switches will induce high overhead in the control plane of SDN. Therefore, a passive flow table collecting method with zero control plane overhead is proposed to address this problem. After finding flows’ real forwarding paths, FlowTrace uses a new measurement method to measure the latencies of different flows. Results of experiments conducted in Mininet indicate that FlowTrace can correctly find flow paths and accurately measure the latencies of flows in different priority classes.     

并行分组交换中负载平衡充要条件的分析

负载平衡是并行分组交换(PPS-parallelpacketswitch)结构的一项关键技术,该文对带缓存PPS结构中负载平衡的充要条件进行了研究。首先基于守恒原理建立流体并行交换的流量守恒方程,并分析流体并行交换中负载平衡的充要条件,在此基础上建立PPS的流量守恒方程,并定义PPS负载平衡,分析PPS负载平衡算法应该满足的充要条件。     

Scalable services via egress admission control

Allocating resources for multimedia traffic flows with real-time performance requirements is an important challenge for future packet networks. However, in large-scale networks, individually managing each traffic flow on each of its traversed routers has fundamental scalability limitations, in both the control plane's requirements for signaling, state management, and admission control, and the data plane's requirements for per-flow scheduling mechanisms. In this paper, we develop a scalable architecture and algorithm for quality-of-service management termed egress admission control. In our approach, resource management and admission control are performed only at egress routers, without any coordination among backbone nodes or per-flow management. Our key technique is to develop a framework for admission control under a general “black box” model, which allows for cross traffic that cannot be directly measured, and scheduling policies that may be ill-described across many network nodes. By monitoring and controlling egress routers' class-based arrival and service envelopes, we show how network services can be provisioned via scalable control at the network edge. We illustrate the performance of our approach with a set of simulation experiments using highly bursty traffic flows and find that despite our use of distributed admission control, our approach is able to accurately control the system's admissible region under a wide range of conditions     

Refined assured forwarding framework for differentiated services architecture

This paper presents a new refined assured forwarding (RAF) framework for improving the performance of DiffServ architecture where heterogeneous traffic flows share the same aggregate class. The new framework requires minimal modification to existing DiffServ routers by adding a second layer of classification of flows based on their average packet sizes and using Weighted Fair Queueing for flow scheduling. The efficiency of the new architecture in enhancing the performance of DiffServ is demonstrated by simulation results for delay, packet delivery, throughput, and packet loss, under different traffic scenarios.     

At the core of IP networks: link-state routing protocols

As end users and managers of both corporate and Internet service provider (ISP) IP networks, we ask a lot of those devices called routers. They must be reliable and easy to manage, while supporting a variety of LAN and WAN interfaces at a reasonable price. They must forward hundreds of thousands or even millions of packets per second. For each packet, this means the router receives it, extracts the destination address contained in the header, performs a lookup in a local routing table, finds the best match, and then transmits the packet to the next-hop router. The router may even be configured to examine additional fields in the packet and, based on this analysis, decide whether to place the packet in a high-priority transmission queue for expedited service. The author discusses link-state routing protocols     

AVMON: Optimal and Scalable Discovery of Consistent Availability Monitoring Overlays for Distributed Systems

This paper proposes to build overlays that help in monitoring of long-term availability histories of hosts, with a focus on large-scale distributed settings where hosts may be selfish or colluding. Concretely, we target the important problems of selection and discovery of an availability monitoring overlay. We motivate six significant goals - firstly, consistency, verifiability, and randomness, in selecting availability monitors of nodes, so as to be probabilistically resilient to selfish and colluding nodes. The next three goals are discoverability, load-balancing, and scalability in finding these monitors. We present AVMON, the first availability monitoring overlay to satisfy these six requirements. Our core algorithmic contribution is a range of protocols for discovering the availability monitoring overlay scalably and efficiently, given any arbitrary monitor selection scheme that is consistent and verifiable. We mathematically analyze the performance of AVMON's discovery protocols w.r.t. scalability and discovery time of monitors. Most interestingly, we are able to derive optimal (and practical) variants of AVMON, that minimize different combinations of memory, bandwidth, computation, and monitor discovery time. Finally, our extensive experimental evaluations using three types of availability traces - synthetic, from PlanetLab, and from the Overnet p2p system - demonstrate AVMON's practicality in a variety of distributed systems.     

DAFT:一种OpenFlow大规模流表区分存储与加速查找架构

软件定义网络作为一种数据转发与逻辑控制相解耦的创新网络范式,当采用OpenFlow协议进行大规模部署时,其数据平面的流表规模急剧增大,对OpenFlow交换机的流表存储资源和分组转发性能提出了严峻的挑战.对此,本文构建了一种OpenFlow大规模流表区分存储与加速查找架构DAFT.该架构根据流量分布特性将OpenFlow网络流区分为重要流和次要流,进而采用TCAM和SRAM分别存储其标识字段,并采用DRAM单独存储其内容字段,有效缓解OpenFlow流表存储资源紧张问题.针对重要流/次要流区分问题,在分析传统大象流/老鼠流区分方法的基础上,基于OpenFlow网络流的包成批特性,提出活跃流/空闲流区分方法,以提高TCAM命中率.针对SRAM流表查找性能瓶颈,利用掩码访问不均匀的特点,采用"往前移1"启发法自适应调整掩码顺序,以减少后续数据包的掩码失败探测次数;利用掩码探测多数会失败的特点,借助计数型布鲁姆过滤器预测元组查找失败结果,从而绕过对应的子流表遍历过程.最后,借助骨干网络流量样本,对本文所提DAFT流表架构的查找性能进行实验评估.实验结果表明:DAFT流表架构的TCAM命中率、SRAM平均查找长度和平均流表访问时间均明显优于传统的大象流/老鼠流架构,且稳定性强,有效提升了OpenFlow交换机的分组转发性能.     

面向CCN的路由转发方案设计

CCN的数据转发具有不确定性和盲目性,降低了转发效率,为了提高CCN的路由转发效率和内容命中率,提出了一种构建内容名称路由表的路由转发方案。CCN路由进行内容名称前缀信息交互,内容名称前缀信息提供构建内容名称路由表所需的名称前缀信息和路由方案,CCN路由时刻维护更新路由内的内容路由表,为兴趣分组提供准确、可靠的路由方案,这样可降低CCN转发过程中的盲目性,提高转发效率。     

基于抽样流记录的RTT估计

往返时延(RTT)是网络测量中的一个重要测度,是刻画网络性能的重要指标。传统的RTT测量都是基于报文的,需要专门的主动或被动测量平台的支持。提出一种新的 RT T 估计方法,仅使用现有路由器设备提供的流记录,不需要额外的网络测量设施。通过对 TCP 块状流传输特性的分析,分别建立了当套接字缓冲区长度与带宽延迟积BDP相对较小、较大和相近这3种情况下的RTT估计模型。实验结果表明,这些模型都能很好地完成RTT估计。同时,由于在估计当中只使用了流持续时间和总报文两个变量,因此,该方法同样适用于以抽样流记录为输入的环境,能够有效地应用于现有的大规模主干网环境的网络检测与管理。     

一种采用流切割实现报文保序的负载均衡算法

网络链路过载或链路失效时,使用负载均衡技术可以避免网络发生拥塞。负载的分派粒度决定了负载均衡系统的均衡性能。分派粒度越细,均衡效果越理想。基于包水平粒度的负载分派可以实现理想的均衡性能,但是会造成同一TCP业务流中报文乱序;基于流水平的分派可以保证报文不乱序,但均衡效果不理想。提出了按照报文段粒度分派负载的FSLB算法。仿真实验表明,该算法可避免报文乱序并能达到较理想的均衡效果。     

随机早期检测算法公平性的改进

随机早期检测（Random Early Detection,RED）算法是广泛应用于路由器中的IP层的拥塞控制算法之一,它的主要思想是在拥塞发生以前通过一定的丢失概率随机丢弃一部分分组,端系统通过收到重复的ACK（Acknowledgement）报文段感知网络即将发生或已经发生网络拥塞,来降低自己的发送速率,从而达到实现网络拥塞控制的目的,但该算法在应用中仍有不足,特别是RED算法在多个流竞争带宽的情况下,不能保证各流获得相对公平的带宽分配。针对于此,在RED算法的基础上设计了一种限制高带宽流算法,以提高的网络公平性。     

Edge-limited scalable QoS flow set-up

Although the Differentiated Services architecture supports scalable packet forwarding based on aggregate flows, the detailed procedure of Quality of Service (QoS) flow set-up within this architecture has not been well established. In this paper we explore the possibility of a scalable QoS flow set-up using a sink-tree paradigm. The paradigm initially constructs a sink tree at each egress edge router using network topology and bandwidth information provided by a QoS extended version of Open Shortest Path First (OSPF), which is a widely used link-state routing protocol. Our sink-tree paradigm dynamically reallocates network bandwidths online according to traffic demands. As a consequence, our paradigm easily supports QoS routing, resource allocation, and admission control at ingress edge routers without consulting core routers in a way that the QoS flow set-up time and overhead are minimized. Simulation results are very encouraging in that the proposed methodology requires significantly less communication overhead in setting up QoS flows compared to the traditional per-flow signaling-based methodology while still maintaining high resource utilization.     

一种基于预测与动态调整负载因子的SDN流表优化算法

通过对SDN流表更新的研究,发现了网络流量高峰期流表更新不及时的问题。提出了一种基于预测与动态调整负载因子的SDN流表优化算法。算法首先收集每个单位时间内的新增流条目,然后用二次移动平均算法对收集的历史数据进行分析,并估计下一个单位时间内新增的流条目,最终根据负载因子动态调整交换机流表中流条目的停滞超时时间。实验结果表明,该算法提高了流表匹配率和数据成功转发率,增加了活动流表项的数量。