Conference Paper

Tackling Bufferbloat in Capacity-limited Networks

Authors:
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Over-provisioned network buffers, often at the In- ternet edge, induce large queuing delay and high latency; this issue is known as Bufferbloat. In response to this, a set of recently proposed Active Queue Management (AQM) algorithms attempt to reduce standing queues, while maintaining the bot- tleneck utilisation at an acceptable level. This paper assesses the performance of two AQM schemes (CoDel and FQ-CoDel) over capacity-limited networks with large Round-Trip Time (RTT). In such settings, these AQM schemes have difficulty controlling the buffering level, resulting in both momentarily high queuing delay and low bottleneck utilisation, even if the methods are claimed to be insensitive to link rates and round-trip delays. We explore this issue and show that it is possible to adapt the parameterisation of CoDel and FQ-CoDel to offer a higher bottleneck utilisation while maintaining a low queuing delay. We present experiments over an emulated test bed and a satellite network to confirm that our new parameterisation improves the download time of moderate-size files and reduces the latency for capacity-limited and large-RTT networks.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

... -The Lower Effort (LE) PHB [26] is currently being specified by the IETF. This traffic is known also known "scavenger" and is only sent if there is no other traffic awaiting transmission at the network interface. ...
... Active queue management (AQM) [26] offers a solution to this problem of sharing the bottleneck buffer without incurring unnecessary delay. New AQM methods such as Flow-Queue-Controlled-Delay (FQ-Codel) [28] support flow differentiation. ...
... -The Lower Effort (LE) PHB [26] is currently being specified by the IETF. This traffic is known also known "scavenger" and is only sent if there is no other traffic awaiting transmission at the network interface. ...
... Active queue management (AQM) [26] offers a solution to this problem of sharing the bottleneck buffer without incurring unnecessary delay. New AQM methods such as Flow-Queue-Controlled-Delay (FQ-Codel) [28] support flow differentiation. ...
Article
Full-text available
Internet Access is regarded by many as a key publicservice in modern societies. The UK Government is pursuing aDigital by Default agenda as part of its digital communicationsinfrastructure strategy to make Internet connectivity availableto all. While it has seen success in enabling connectivity to themajority of the population, the geographical location of the final5% of hard-to-reach premises pose technical challenges,preventing traditional broadband services from being costeffective. In most remote and hard-to-reach areas, satellitebroadband can provide connectivity, however, the presentdesign of service offerings is unable to provide affordable or“Free” Internet service models combined with acceptableperformance. One alternative is to offer lower-effort (LE) accessto Internet services to provide "Free Internet". In contrast toterrestrial networks, providing LE services over a satellitenetwork requires optimal cross-layer traffic engineering (TE) tooffer acceptable performance. This paper presents anexperimental service platform called Rural-PAWS (Rural-Public Access Wi-Fi Services). This was designed to support freehigh-speed access to government services, with basic access totraditional Internet services over satellite. The Rural-PAWSmodel uses a prototype multilevel service model thatdistinguishes Government Digital Services from LE traffic. Theplatform was deployed in a 12 months study over satellite across8 pilot sites. This study sought to understand actual servicerequirements, user perceived performance of web access oversatellite-based LE service, while investigating technicalchallenges to enable deeper understanding of whether it istechnically viable to offer such services to hard-to-reachcommunities in the UK. The experimental results show that anLE service over satellite network can be provisioned, providedthat (i) the web browser parameters are optimised for satellitenetworks and (ii) suitable application layer protocols are enabled over satellite networks.
... Each consecutive packet drop increases n drop + = 1 and update µ value to µ √ n drop . If δ i < τ , CoDel enters in the non-dropping state, being the drop count (n drop ) reset to 1 and µ to λ [20]. The Flow Queue CoDel (FQ-CoDel) variation uses the same CoDel approach to control the packet sojourn time in queue. ...
... Related to AQMs, authors in [20] evaluated CoDel and FQ-CoDel over networks with limited transmission capacity and high RTTs. In order to reduce latency, they optimized the CoDel parameters for a specific scenario. ...
Article
Full-text available
Heterogeneous wireless networks have evolved to reach application requirements for low latency and high throughput on Internet access. Recent studies have improved network performance employing the Multipath TCP, which aggregates flows from heterogeneous wireless interfaces in a single connection. Although existing proposals are powerful, coupled congestion control algorithms are currently limited because of the high variation in path delays, bandwidth and loss rate, typical from heterogeneous wireless networks, even more over concurrent multipath transmissions. These transmissions experience bufferbloat, i.e., high delays caused by long queues. Hence, to cope with the current limitations, this work presents CoDel-LIFO, a new active queue management (AQM) discipline to reduce the dropped packet ratio in the Multipath TCP congestion control mechanism. Differently from other approaches, CoDel-LIFO gives priority to the most recent packets, being then promising. This paper provides a detailed simulation analysis over congestion control algorithms by comparing CoDel-LIFO to CoDel and DropTail disciplines. Results indicate that CoDel-LIFO reduces queue drops, diminishing the impact on congestion control; improving substantially the goodput; and keeping RTT low.
... CoDel has been shown to have auto-tuning issues and its performance is sensitive to the traffic load [7]. Also, its default 5 ms of maximum allowed queuing delay can be damaging for low-speed bottlenecks [8] and its interval value is based on the assumption that the Round Trip Time (RTT) is 100 ms [9]. On the other hand, PIE has been shown to be less sensitive to traffic loads [7], its default target delay of 20 ms should be less problematic with low capacity bottlenecks, and it does not make assumptions on the RTT. ...
... Indeed, in [7], CoDel has been shown to have autotuning issues and its performance are sensitive to the load of traffic. Also, 5 ms of maximum allowed queuing delay can be damaging for bottlenecks of 2 Mbps [8] and its interval value is based on the assumption that the RTT is 100 ms [9], which is not the case for rural broadband networks. On the contrary, our algorithm clearly improves the performance of PIE when the RTT is higher than 300 ms for various types of traffic and does not affect the performance of PIE for lower RTTs. ...
Technical Report
Full-text available
Bufferbloat is excessive latency due to over-provisioned network buffers. PIE and CoDel are two recently proposed Active Queue Management (AQM) algorithms, designed to tackle bufferbloat by lowering the queuing delay without degrading the bottleneck utilization. PIE uses a proportional integral controller to maintain the average queuing delay at a desired level; however, large Round Trip Times (RTT) result in large spikes in queuing delays, which induce high dropping probability and low utilization. To deal with this problem, we propose Maximum and Average queuing Delay with PIE (MADPIE). Loosely based on the drop policy used by CoDel to keep queuing delay bounded, MADPIE is a simple extension to PIE that adds deterministic packet drops at controlled intervals. By means of simulations, we observe that our proposed change does not affect PIE's performance when RTT < 100 ms. The deterministic drops are more dominant when the RTT increases, which results in lower maximum queuing delays and better performance for VoIP traffic and small file downloads, with no major impact on bulk transfers.
... Therefore, a 'safety factor' is applied to the average RTT between users and CDNs, which has to allow for the statistical distribution of RTTs to CDNs, particularly for users in rural areas [KKFR15], who will be further from the nearest CDN and who are also likely to have least bandwidth and therefore be least willing to see it eaten by under-utilization. The safety factor also has to allow for flows between clients and servers other than in CDNs. ...
Preprint
This report gives the reasoning for the parameter settings of the reference Linux implementation of the PI$^2$ AQM, focusing initially on the target queue delay.
... These algorithms significantly improve the stability of queue size. The newly algorithms CoDeL and FQ-CoDeL proposed in [23][24] aim to reduce the delay. These schemes can achieve a higher bottleneck utilization while maintaining a low queuing delay over capacity-limited networks with large Round-Trip Time (RTT). ...
Article
Most active queue management algorithms manage network congestion based on the size of the queue but ignore the network environment which makes queue size change. It seriously affects the response speed of the algorithm. In this paper, a new AQM algorithm named CT-AQM (Change Trend-Adaptive Queue Management) is proposed. CT-AQM predicts the change trend of queue size in the soon future based on the change rate of queue size and the network environment, and optimizes its dropping function. Simulation results indicate that CT-AQM scheme has a significant improvement in loss-rate and throughput.
... Studies have shown that CoDel has self-adaptation issues, and its performance is sensitive to the appropriate setting of target and interval [23][24][25][26]. It has also been noted that CoDel's control law fails to adapt and control the queue length when a large number of unresponsive UDP flows coexist with responsive TCP flows [22]. ...
Article
Full-text available
In this study, a modified fluid model is proposed to understand the design of controlled delay (CoDel) algorithm for active queue management (AQM) and analyse its sensitivity to parameter settings. CoDel significantly differs from other AQM algorithms because it operates at the head of the queue and adopts a deterministic packet drop strategy, unlike other algorithms that operate at the tail and adopt a probabilistic packet drop strategy. The correctness of the proposed fluid model is verified by comparing its results with those obtained from ns-2. Subsequently, using the model developed in this study, the authors analyse the performance of CoDel algorithm by changing its internal parameters and modifying its control law. They highlight the role of the internal parameters and control law on the ability of the CoDel algorithm to control queue delay. Their analysis shows that the CoDel algorithm is sensitive to its parameter settings and that its control law requires minor modifications to gain a better control over the queue delay.
... PSPAT supports a class of Scheduling Algorithms (such as DRR [5], WF 2 Q+ [6], QFQ [7]) that provide isolation and provable, tight service guarantees with arbitrary workloads. These cannot be compared with: (1) queue management schemes such as FQ_CODEL [8], or OS features such as PFIFO_FAST or multiqueue NICs, often incorrectly called "schedulers", that do not provide reasonable isolation or service guarantees; (2) heuristic solutions based on collections of shapers reconfigured on coarse timescales, that also cannot guarantee fair short term rate allocation; or (3) more ambitious systems that try to do resource allocation for an entire rack or datacenter [9], which, due to the complexity of the problem they address, cannot give guarantees at high link utilization. More details are given in Section 7. ...
Article
Full-text available
Tenants in a cloud environment run services, such as Virtual Network Function instantiations, that may legitimately generate millions of packets per second. The hosting platform, hence, needs robust packet scheduling mechanisms that support these rates and, at the same time, provide isolation and dependable service guarantees under all load conditions. Current hardware or software packet scheduling solutions fail to meet all these requirements, most commonly lacking on either performance or guarantees. In this paper we propose an architecture, called PSPAT to build efficient and robust software packet schedulers suitable to high speed, highly concurrent environments. PSPAT decouples clients, scheduler and device driver through lock-free mailboxes, thus removing lock contention, providing opportunities to parallelize operation, and achieving high and dependable performance even under overload. We describe the operation of our system, discuss implementation and system issues, provide analytical bounds on the service guarantees of PSPAT, and validate the behavior of its Linux implementation even at high link utilization, comparing it with current hardware and software solutions. Our prototype can make over 28 million scheduling decisions per second, and keep latency low, even with tens of concurrent clients running on a multi-core, multi-socket system.
... A statistical analysis for calculating exact dropping probability (EDP) is developed to control the instantaneous queue length in [11]. Delay based AQM algorithms, namely, controlled delay (CoDel) and flow-queuing controlled delay (FQ-CoDel) have been presented in [12]. ...
Article
Full-text available
Recently, a dynamic adaptive queue management with random dropping (AQMRD) scheme has been developed to capture the time-dependent variation of average queue size by incorporating the rate of change of average queue size as a parameter. A major issue with AQMRD is the choice of parameters. In this paper, a novel online stochastic approximation based optimization scheme is proposed to dynamically tune the parameters of AQMRD and which is also applicable for other active queue management (AQM) algorithms. Our optimization scheme significantly improves the throughput, average queue size, and loss-rate in relation to other AQM schemes.
... Control processes related to the MAC layer are indeed closely related to the characteristics of the medium used to transmit the data. Some related work proposed specific controllers for satellite communications [38][39][40]. However, the dynamically controlled parameters could be located at the satellite platform level, at the access gateway, or at the network level. ...
Article
This paper provides an assessment study on the virtualization of a DVB-S2/DVB-RCS2 satellite ground infrastructure and proposes a framework, named SatCloudRAN, which aims to ease the integration of satellite components in forthcoming 5G systems. Special attention is given to the design of SatCloudRAN by considering the split and placement of virtualized and non-virtualized functions while taking into account the characteristics of the transport links connecting both type of functions. We assess how virtualization and softwarization technologies, namely Network Function Virtualization (NFV) and Software Defined Networking (SDN), can deliver part of the satellite gateway functionalities as Virtual Network Functions and achieve a flexible and programmable control and management of satellite infrastructure. Under the NFV paradigm, building VNF blocks that compose a satellite gateway have been identified, and their interaction exhibited. This paper also gives insights on how the SatCloudRAN approach can allow operators to provide SDN-based: (1) bandwidth on demand, (2) dynamic QoS, and (3) satellite gateway diversity.
Conference Paper
Reinforcement learning has gone through an enormous evolution in the past ten years. It's practical applicability has been demonstrated through several use cases in various fields from robotics to process automation. In this paper, we examine how the tools of deep Q-learning can be used in an AQM algorithm to reduce queuing delay and ensure good link utilization at the same time. The proposed method called RL-AQM has the advantage that it is less prone to the good parameterization and can automatically adapt to new network conditions. The prototype implementation based on OpenAI Gym and NS-3 network simulator has thoroughly been evaluated under various settings focusing on three aspects: the convergence time of learning process, the performance of pre-trained models compared to PIE AQM and generalization ability. We have demonstrated that RL-AQM achieves comparable utilization to PIE AQM but results in much smaller queueing delays. Finally, the pre-trained models have good generalization abilities, enabling to use a pre-trained model in network settings that differ in bandwidth and/or RTT from the one used during the pre-training phase.
Chapter
Novel interactive applications require small end-to-end latency for providing end users with good Quality of Experience. To prevent the bufferbloat problem causing increased delay, various AQM solutions have recently emerged. One of the most widely adopted method is called CoDel that detects the increased delay by maintaining the per packet sojourn times and compering them to a desired target delay. Accordingly, congestion situation is indicated when the sojourn time exceeds the target delay. CoDel applies a drop-based feedback to notify the responsive sources about congestion only when the deviation from the target delay is permanent. Dropping packets is then compensated by packet re-transmissions in case of TCP which worsens utilization. In this paper, we propose an ECN-enhanced CoDel that distinguishes between low and high levels of congestion, and ECN-marks or drops the packets, respectively. The performance of the proposed method is analyzed in thorough NS-3 simulations with variable number of flows. The proposed ECN-marking reduces the number of packets re-transmissions and provides similarly good characteristics including convergence time and fairness to the original drop-based CoDel.
Article
Bufferbloat is excessive delay due to the accumulation of packets in a router’s oversized queues. CoDel and PIE are two recent Active Queue Management (AQM) algorithms that have been proposed to address bufferbloat by reducing the queuing delay while trying to maintain a high bottleneck utilization. This paper fills a gap by outlining what are the operating ranges, that is the network characteristics (in terms of round-trip times and bottleneck capacity), for which these algorithms achieve their design goals. This new approach to the problem lets us identify deployment scenarios where both AQM schemes result in poor performance when used with default parameters. Because PIE and CoDel have been proposed with RED’s deployment issues in mind, it was essential to evaluate to what extent we can tune them to achieve various trade-offs and let them control the queuing delay outside their default operating range. We find that, by appropriate tuning (1) the amount of buffering can easily be controlled with PIE, (2) the Round Trip Time (RTT) sensitivity of CoDel can be reduced. Also, we observe there is more correlation between the congestion level, the achieved queuing delay and the targeted delay with CoDel than with PIE. This paper therefore concludes there is no single overall best AQM scheme, as each scheme proposes a specific trade-off.
Conference Paper
Full-text available
Bufferbloat is a phenomenon where excess buffers in the network cause high latency and jitter. As more and more interactive applications (e.g. voice over IP, real time video conferencing and financial transactions) run in the Internet, high latency and jitter degrade application performance. There is a pressing need to design intelligent queue management schemes that can control latency and jitter; and hence provide desirable quality of service to users. We present here a lightweight design, PIE (Proportional Integral controller Enhanced), that can effectively control the average queueing latency to a reference value. The design does not require per-packet extra processing, so it incurs very small overhead and is simple to implement in both hardware and software. In addition, the design parameters are self-tuning, and hence PIE is robust and optimized for various network scenarios. Simulation results, theoretical analysis and Linux testbed results show that PIE can ensure low latency and achieve high link utilization under various congestion situations.
Article
Full-text available
This article discusses the danger of a growing digital divide between rural and other areas. It presents broadband as increasingly necessary for the delivery of information, health, education, business, social security, public and leisure services. Access to broadband has become vital for rural communities to participate in a progressively digital economy and to overcome problems of physical and social isolation. Yet rural areas are among those most excluded from fast broadband developments. Although this is partly due to technological/economic barriers in reaching more remote locations, even where technology is available, adoption can still be low in rural areas. This article explores the problems of providing broadband in rural Britain, considers various technological approaches and concludes with key development areas for policy and government.
Article
Full-text available
The authors present random early detection (RED) gateways for congestion avoidance in packet-switched networks. The gateway detects incipient congestion by computing the average queue size. The gateway could notify connections of congestion either by dropping packets arriving at the gateway or by setting a bit in packet headers. When the average queue size exceeds a present threshold, the gateway drops or marks each arriving packet with a certain probability, where the exact probability is a function of the average queue size. RED gateways keep the average queue size low while allowing occasional bursts of packets in the queue. During congestion, the probability that the gateway notifies a particular connection to reduce its window is roughly proportional to that connection's share of the bandwidth through the gateway. RED gateways are designed to accompany a transport-layer congestion control protocol such as TCP. The RED gateway has no bias against bursty traffic and avoids the global synchronization of many connections decreasing their window at the same time. Simulations of a TCP/IP network are used to illustrate the performance of RED gateways
Conference Paper
In this paper we compare the three Active Queue Management (AQMs) Adaptive Random Early Detection (ARED), Controlled Delay (CoDel) and Proportional Integral controller Enhanced (PIE) in static as well as dynamic scenarios. We find significant issues when these algorithms are used for big Round Trip Times (RTTs) as well as a significant utilization decrease when used for high bandwidth links. When used for low and medium sized links, CoDel, PIE and ARED are suitable alike, but for corner scenarios clear recommendations can be given.
Conference Paper
In the past, networks have been mainly optimized for good system throughput but recently achieving low delay has also gained notable traction. Active Queue Management (AQM) has long been recognized necessary for operating Internet routers with shorter standing queues but only limited deployment has occurred. The recent interest in AQM has resulted in new AQM proposals. In this paper we evaluate CoDel (Controlled Delay) and PIE (Proportional Integral controller Enhanced), both being new AQM proposals, and compare the performance against an aggressive RED (Random Early Detection) variant called HRED (Harsh RED). We focus on AQM behavior during load transients typically occurring at the network edge with the common traffic types of today such as Web transactions. We discover that CoDel auto-tuning does not scale well with the load. With the high-end delays experienced, HRED is better than PIE and CoDel when more than a few simultaneous flows share the bottleneck.
Article
The article aims to provide part of the buffer bloat solution, proposing an innovative approach to AQM suitable for today's Internet called CoDel. Packet networks require buffers to absorb short-term arrival rate fluctuations. Although essential to the operation of packet networks, buffers tend to fill up and remain full at congested links, contributing to excessive traffic delay and losing the ability to perform their intended function of absorbing bursts. The Internet has been saved from disaster by a constant increase in link rates and by usage patterns. Over the past decade, evidence has accumulated that this whistling in the dark cannot continue without severely impacting Internet usage. Developing effective active queue management has been hampered by misconceptions about the cause and meaning of queues. Network buffers exist to absorb the packet bursts that occur naturally in statistically multiplexed networks.
Article
This paper presents Random Early Detection (RED) gateways for congestion avoidance in packet-switched networks. The gateway detects incipient congestion by computing the average queue size. The gateway could notify connections of congestion either by dropping packets arriving at the gateway or by setting a bit in packet headers. When the average queue size exceeds a preset threshold, the gateway drops or marks each arriving packet with a certain probability, where the exact probability is a function of the average queue size. RED gateways keep the average queue size low while allowing occasional bursts of packets in the queue. During congestion, the probability that the gateway notifies a particular connection to reduce its window is roughly proportional to that connection's share of the bandwidth through the gateway. RED gateways are designed to accompany a transport-layer congestion control protocol such as TCP. The RED gateway has no bias against bursty traffic and avoids the global synchronization of many connections decreasing their window at the same time. Simulations of a TCP/IP network are used to illustrate the performance of RED gateways.
Article
We have conflated "speed" with "band width." As Stuart Chesire wrote in "It's the Latency, Stupid" (http://rescomp.stanford.edu/~cheshire/rants/Latency.html), "Making more bandwidth is easy. Once you have bad latency, you're stuck with it." Bufferbloat is the existence of excessively large (bloated) buffers in systems, particularly network communication systems. Bufferbloat is now (almost?) everywhere. Today's routers, switches, gateways, broad band gear, and so on have bloated buffer sizes to where we often measure latency in seconds, rather than microseconds or milliseconds.
Flowqueue-codel (work in progress)
  • T Hoeiland-Joergensen
  • P Mckenney
  • D Taht
  • J Gettys
  • E Dumazet
T. Hoeiland-Joergensen, P. McKenney, D. Taht, J. Gettys, and E. Dumazet, " Flowqueue-codel (work in progress), " 2013, IETF.
A PIE-Based AQM for DOCSIS Cable Modems (work in progress)
  • G White
  • R Pan
G. White and R. Pan, " A PIE-Based AQM for DOCSIS Cable Modems (work in progress), " 2014, IETF.
Controlled delay active queue management (work in progress)
  • K Nichols
  • V Jacobson
  • A Mcgregor
  • J Iyengar
K. Nichols, V. Jacobson, A. McGregor, and J. Iyengar, " Controlled delay active queue management (work in progress), " 2014, IETF.