Multiple Bottleneck Links Research Articles

Machine learning for measurement-based bandwidth estimation

The dispersion that arises when packets traverse a network carries information that can reveal relevant network characteristics. Using a fluid-flow model of a bottleneck link with first-in first-out multiplexing, accepted probing tools measure the packet dispersion to estimate the available bandwidth, i.e., the residual capacity that is left over by other traffic. Difficulties arise, however, if the dispersion is distorted compared to the model, e.g., by non-fluid traffic, multiple bottlenecks, clustering of packets due to interrupt coalescing, and inaccurate time-stamping in general. It is recognized that modeling these effects is cumbersome if not intractable. This motivates us to explore the use of machine learning in bandwidth estimation. We train a neural network using vectors of the packet dispersion that is characteristic of the available bandwidth. Our testing results reveal that even a shallow neural network identifies the available bandwidth with high precision. We also apply the neural network under a variety of notoriously difficult conditions that have not been included in the training, such as randomly generated networks with the multiple bottleneck links and heavy cross traffic burstiness. Compared to two state-of-the-art model-based techniques as well as a recent machine learning-based technique (Yin et al., 2016), our neural network approach shows improved performance. Further, our neural network can effectively control the estimation procedure in an iterative implementation. We also evaluate our method with other supervised machine learning techniques.

Modeling TCP Incast Issue in Data Center Networks and an Adaptive Application-Layer Solution

In data centers, the transmission control protocol (TCP) incast causes catastrophic goodput degradation to applications with a many-to-one traffic pattern. In this paper, we intend to tame incast at the receiver-side application. Towards this goal, we first develop an analytical model that formulates the incast probability as a function of connection variables and network environment settings. We combine the model with the optimization theory and derive some insights into minimizing the incast probability through tuning connection variables related to applications. Then, enlightened by the analytical results, we propose an adaptive application-layer solution to the TCP incast. The solution equally allocates advertised windows to concurrent connections, and dynamically adapts the number of concurrent connections to the varying conditions. Simulation results show that our solution consistently eludes incast and achieves high goodput in various scenarios including the ones with multiple bottleneck links and background TCP traffic.

선박 내 유선망에서 차등화 서비스 지원을 위한 패킷 폐기 기술

선내 유선 네트워크에서는 선내의 각종 센서와 제어기를 자율적으로 구성관리하고 원격제어를 제공하는 기능을 수행한다. 이러한 선박 내 유선 통신망에서 차등화서비스가 지원된다면, 우선순위가 높은 실시간 플로우는 최소전송률을 보장 받고 빠르게 전송되어, 지능적인 선박 통신망을 구현할 수 있다. 본 논문에서는 이러한 선박 내 유선 네트워크에서 차등화 서비스를 지원하기 위해, 기존 RIO (RED In and Out) 방식보다 우수한 패킷 폐기 기술을 제안한다. 제안한 패킷 폐기기술은 개별 플로우별로 관리하지 않고 특정한 기준에 따라 몇 개의 플로우 그룹으로 나누어, 그룹별 관리를 통해서 동일 클래스 내 플로우 간 공평성을 제공한다. 시뮬레이션 결과에서는 RIO 방식보다 링크 이용률은 조금 감소하나, 다중 병목 구간을 경유하는 플로우들이 RIO 방식보다 더 많은 전송률을 보장 받음을 알 수 있다. An wired ship area network has functionality of remote control and autonomous management of various sensors and instruments embedded or boarded in a ship. For such environment, the DiffServ (Differentiated Services) realizes that the high-speed real-time flow with the higher priority has the guaranteed minimum data rate and is delivered faster. As a result of this DiffServ effect, the intelligent Ship Area Networks can be implemented. In this paper, an packet drop technique is proposed to outperform the previous RIO (RED In and Out) drop mechanism for DiffServ in ship area networks. the proposed packet drop technique does not manage the individual flows and divides them into several flow groups according to a criterion. And it guarantees the fairness between individual flows in the same QoS class through the group-based control. In simulation results of the proposed packet drop technique, the link utilization decreases than RIO. But it guarantees more data rates to DiffServ flows passing multiple bottleneck links.

Stability analysis of multiple-bottleneck networks

A TCP/RED (Transmission Control Protocol/Random Early Detection) system with multiple-bottleneck links could be unstable even if its system parameters are set the same as those in a stable single-bottleneck system [D. Bauso, L. Giarre, G. Neglia, Active queue management stability in multiple bottleneck networks, IEEE ICC’04, vol. 4, June 2004, pp. 2267–2271]. In this paper, we study the stability of more general AIMD (Additive Increase and Multiplicative Decrease)/RED system with multiple bottlenecks that may incur non-negligible packet losses. We develop a general mathematical model to analyze network stability for both delay-free and delayed AIMD/RED systems. Sufficient conditions for the asymptotic stability of multiple-bottleneck systems with heterogeneous delays are derived by appealing to Lyapunov stability theory with Lyapunov–Razumikhin conditions, and these conditions can be easily assessed by using LMI (Linear Matrix Inequality) Toolbox. Numerical results with Matlab and simulation results with NS-2 are given to validate the analytical results.

Self-tuning TCP traffic controller using gain margin specification

The paper proposes a self-tuning active queue management (AQM) controller for Internet protocol (IP) routers supporting transmission control protocol (TCP) traffic in the Internet. Classical control theory is applied in the design to meet the frequency-domain-specification gain margin. The user can assign a proper gain margin to achieve good AQM performance. Simple equations for obtaining the AQM controller are derived. The designed AQM controller self-tunes, only when changes in Internet traffic have drifted the traffic monitoring parameter outside its specified interval, otherwise the self-tuning AQM controller remains unchanged. The OPNET® simulations used by the authors in single and in multiple bottleneck links demonstrate that, with the use of a self-tuning AQM controller in the routers, this AQM control system adapts very well to great fluctuations in the TCP traffic, thus providing the network with good stability robustness.

Lightweight proactive queue management

The quest for better resource control has been the driving force behind Active Queue Management (AQM) research. Random Early Detection (RED), the defacto standard and its variants have been proposed as simple solutions to the AQM problem. These approaches, however, are known to suffer from problems like parameter sensitivity and inability to capture input traffic load fluctuations accurately, thereby resulting in instability. This paper presents a proactive queue management algorithm called PAQMAN that captures input traffic load fluctuations accurately and regulates the queue size around the desirable level. PAQMAN draws from the predictability in the underlying traffic by employing the Recursive Least Squares (RLS) algorithm to forecast the average queue size over the next prediction interval using the average queue size information of the past intervals. The packet drop probability is then computed as a function of this predicted average queue size. The performance of PAQMAN has been evaluated and compared against existing AQM schemes through ns-2 simulations that encompass varying network conditions for networks comprising of single as well as multiple bottleneck links. Simulation results demonstrate that PAQMAN maintains a relatively low queue size, while at the same time achieving high link utilization and low packet loss. Moreover, the computational overhead of PAQMAN is negligible (lightweight) which further justifies its use.

Public review for LTCP

LTCP is a new modification to TCP's congestion control to enable it to scale to high-speed networks. In comparison with prior work on high-speed congestion control, the paper explores an interesting design space: what if a single flow acts similarly to K virtual TCPs, where K is dynamically adjusted according to the network's congestion state; K increases in times of under-load to aggressively grab the available bandwidth and decreases in times of congestion to quickly release the bandwidth. The paper elaborates on how to design such a protocol and particularly how to make it fair when multiple flows with a different initial K share the same bottleneckThough I find it interesting to explore such a design, with much prior work on high-speed congestion control, it is unclear what LTCP can add. The paper claims that one of the contributions of LTCP is that "it retains the time-tested AIMD behavior of TCP." Yet, because of the dynamic value of K, LTCP does not really do AIMD. Besides, why is maintaining AIMD important? The paper does not elaborate on this issue; neither does it show that the use of different increase-decrease rule has hurt other high-speed congestion control protocolsOverall, LTCP is a nice piece of research but I would encourage the authors to argue more why their design is important given all prior work on the topic. Further, they should run more extensive simulations that explore multiple bottleneck links, and provide at least a simple analysis that supports the convergence claim.

A distributed utility max–min flow control algorithm

A fair allocation of utility (application-layer performance) is essential in providing QoS (Quality of Service) guarantee. However, there have been few researches in the literature of utility-fair network resource allocation scheme. In this paper, we propose a distributed utility max–min flow control algorithm which accommodates application diversity in that it does not require the concavity of utility functions, and is scalable in that it does not require any per-flow operation in the network. The algorithm is proved to be convergent under the assumption that there exists a single bottleneck link and the communication delay between any two links in the network is bounded. Although the convergence of the algorithm is analyzed only for the case of a single bottleneck link, we show through simulations that the proposed algorithm works as designed for the case of multiple bottleneck links as well.

Adaptive prediction based approach for congestion estimation (APACE) in active queue management

Active Queue Management (AQM) policies provide an early indication of incipient congestion to the sources. In this paper, we propose a new AQM policy called APACE. APACE stands for Adaptive Prediction based Approach for Congestion Estimation in AQM that predicts the instantaneous queue length at a future time instant using adaptive filtering techniques. We compare the performance of APACE with other existing AQM schemes in networks having both single and multiple bottleneck links. We show that APACE is able to control the oscillations in the instantaneous queue. We also demonstrate, through exhaustive simulations, that APACE performs well in terms of link utilization even in networks with multiple bottleneck links. Moreover, APACE is not very sensitive to parameter settings and adapts quickly to changes in traffic.