Results 1  10
of
24
The effects of fairness in buffer sizing
, 2007
"... Abstract. Buffer sizing in Internet routers is a fundamental problem that has major consequences in the design, implementation, and economy of the routers, as well as on the performance observed by the end users. Recently, there have been some seemingly contradictory results on buffer sizing. On the ..."
Abstract

Cited by 10 (2 self)
 Add to MetaCart
Abstract. Buffer sizing in Internet routers is a fundamental problem that has major consequences in the design, implementation, and economy of the routers, as well as on the performance observed by the end users. Recently, there have been some seemingly contradictory results on buffer sizing. On the one hand, Appenzeller et al. show that as a direct consequence of desynchronization of flows in the core of the Internet, buffer sizes in core routers can be significantly reduced without any major degradation in network performance. On the other hand, Raina and Wischik show that such reduction in buffer sizing comes at the cost of synchronization and thus instability in the network. This work unifies these results by studying the effects of fairness in buffer sizing. We show that the main difference arises from the implicit assumption of fairness in packet dropping in the latter result. We demonstrate that desynchronization among flows observed by Appenzeller et al. is caused by unfair packet dropping when a combination of TCPReno and the droptail queue management is used. We also show that bringing fairness in packet dropping will introduce synchronization among flows, and will make the system unstable as predicted by Raina and Wischik. Our analysis suggests that there is an intrinsic tradeoff between fairness in packet drops and desynchronization among TCPReno flows when routers use the droptail queue management. Achieving fairness, desynchronization, small buffer size, and 100 % link utilization at the same time is desirable and feasible yet challenging. The studies in this paper provide insights for further explorations in reaching this goal. 1
On the interaction between TCPlike sources and throughputefficient scheduling policies
, 2007
"... ..."
(Show Context)
A Theory of Load Adjustments and its Implications for Congestion Control
 Journal of Internet Engineering, Klidarithmos Press
, 2007
"... and Multiplicative Decrease (MD) are linear adjustments used extensively in networking. However, their properties are not fully understood. We analyze responsiveness (time for the total load to reach the target load), smoothness (maximal size of the total load oscillations after reaching the target ..."
Abstract

Cited by 6 (2 self)
 Add to MetaCart
and Multiplicative Decrease (MD) are linear adjustments used extensively in networking. However, their properties are not fully understood. We analyze responsiveness (time for the total load to reach the target load), smoothness (maximal size of the total load oscillations after reaching the target load), fairing speed (speed of convergence to equal individual loads) and scalabilities of MAIMD algorithms, which generalize AIMD algorithms via optional inclusion of MI. We prove that an MAIMD can provide faster asymptotic fairing than a less smooth AIMD. Furthermore, we discover that loads under a specific MAIMD converge from any initial state to the same periodic pattern, called a canonical cycle. While imperfectly correlated with smoothness, the canonical cycle reliably predicts the asymptotic fairing speed. We also show that AIMD algorithms offer the best tradeoff between smoothness and responsiveness. Then, we introduce smoothnessresponsiveness diagrams to investigate MAIMD scalabilities. Finally, we discuss implications of the theory for the practice of congestion control. I.
Extending the TCP SteadyState Throughput Equation for Parallel TCP Flows
"... Abstract—We present a simple extension of the wellknown TCP steadystate throughput equation by Padhye et al. which can be used to calculate the throughput of several flows that share an endtoend path at the same time. The value of this extension, which we show to work well with validations using ..."
Abstract

Cited by 4 (1 self)
 Add to MetaCart
(Show Context)
Abstract—We present a simple extension of the wellknown TCP steadystate throughput equation by Padhye et al. which can be used to calculate the throughput of several flows that share an endtoend path at the same time. The value of this extension, which we show to work well with validations using ns2 simulations as well as reallife measurements, is its applicability in practice and its ease of use. I.
Timestep Stochastic Simulation of Computer Networks using Diffusion Approximation
, 1903
"... Timestep stochastic simulation (TSS) is a novel method for generating sample paths of computer networks, with low computation cost independent of packet rates. It has accuracy adequate to evaluate general network and flow configurations, including arbitrary flow start times and durations, droptail ..."
Abstract

Cited by 3 (1 self)
 Add to MetaCart
(Show Context)
Timestep stochastic simulation (TSS) is a novel method for generating sample paths of computer networks, with low computation cost independent of packet rates. It has accuracy adequate to evaluate general network and flow configurations, including arbitrary flow start times and durations, droptail queuing (i.e., does not require RED), and arbitrary statedependent control mechanisms for congestion control and routing. TSS generates the evolution of the system state S(t) on a sample path in time steps of size δ. At each step, S(t+δ) is randomly chosen according to S(t) and the probability distribution P r[S(t + δ)S(t)] obtained using the diffusion approximation. Because packet transmission and reception events are replaced by time steps, TSS generates sample paths at a fraction of the cost of packetlevel simulation. Because TSS generates sample paths, control feedback can be based on sample path metrics, rather than ensemble metrics, thereby accurately capturing the effects of statedependent control mechanisms. 1
Feasibility Aspects of AMP Performance Evaluation in a Fluid Simulation Environment
 in Proc. 3rd MMBnet Workshop
, 2005
"... In order to address the problem of intradomain IP traffic engineering efficiently, we have recently proposed Adaptive MultiPath routing (AMP) as a dynamic algorithmic solution based on a local view of the network. For two real ISP topologies we have demonstrated that our scheme significantly impro ..."
Abstract

Cited by 3 (3 self)
 Add to MetaCart
(Show Context)
In order to address the problem of intradomain IP traffic engineering efficiently, we have recently proposed Adaptive MultiPath routing (AMP) as a dynamic algorithmic solution based on a local view of the network. For two real ISP topologies we have demonstrated that our scheme significantly improves network performance while generating only minimal signaling overhead. However, our performance evaluations have been accomplished by packetlevel simulation, which has turned out to be a rather slow method as soon as realistic network sizes and traffic scenarios are considered, and thus prevents a deeper analysis of the algorithm. Therefore, we are currently developing a simulation environment which is based on the fluid model, i.e., focusing on flow aggregates instead of single packet events and connections, which makes it much faster than packetlevel simulation. In this paper we describe the basic concepts of fluid models and evaluate the feasibility of fluidbased simulation of AMP, particularly concentrating on computational efficiency. 1
An extension of the tcp steadystate throughput equation for parallel flows and its application in multfrc
 Networking, IEEE/ACM Transactions on
, 2011
"... Abstract—In the first part of this paper, we present a simple extension of the wellknown TCP steadystate throughput equation that can be used to calculate the throughput of several flows that share an endtoend path. The value of this extension, which we show to work well with simulations as wel ..."
Abstract

Cited by 2 (2 self)
 Add to MetaCart
(Show Context)
Abstract—In the first part of this paper, we present a simple extension of the wellknown TCP steadystate throughput equation that can be used to calculate the throughput of several flows that share an endtoend path. The value of this extension, which we show to work well with simulations as well as reallife measurements, is its practical applicability. Thus, in the second part of this paper, we present its application in MulTFRC, a TCPfriendly rate control (TFRC)based congestion control mechanism that is fair to a number of parallel TCP flows while maintaining a smoother sending rate than multiple real TFRC flows do. MulTFRC enables its users to prioritize transfers by controlling the fairness among them in an almost arbitrary fashion. Index Terms—Computer networks, modeling, protocols. I.
Performance of TCP on lowbandwidth wireless links with delay spikes
"... We model the goodput of a single TCP source on a wireless link experiencing sudden increases in RTT, i.e., delay spikes. Such spikes trigger spurious timeouts that reduce the TCP goodput. Renewal reward theory is used to derive a straightforward expression for TCP goodput that takes into account lim ..."
Abstract
 Add to MetaCart
(Show Context)
We model the goodput of a single TCP source on a wireless link experiencing sudden increases in RTT, i.e., delay spikes. Such spikes trigger spurious timeouts that reduce the TCP goodput. Renewal reward theory is used to derive a straightforward expression for TCP goodput that takes into account limited sending rates (limited window size), lost packets due to congestion and the delay spike properties such as the average spike duration and distribution of the spike intervals. The basic model is for i.i.d. spike intervals, and correlated spike intervals are modeled by using a modulating background Markov chain. Validation by ns2 simulations shows excellent agreement for lossless scenarios and good accuracy for moderate loss scenarios (for packet loss probabilities less than 5%). Numerical studies have also been performed to assess the impact of different spike interval distributions on TCP performance.
1 Buffer sizes for large multiplexers:
"... Abstract — In large multiplexers with many TCP flows, the aggregate traffic flow behaves predictably; this is a basis for the fluid model of Misra, Gong and Towsley [1] and for a growing literature on fluid models of congestion control. In this paper we argue that different fluid models arise from d ..."
Abstract
 Add to MetaCart
(Show Context)
Abstract — In large multiplexers with many TCP flows, the aggregate traffic flow behaves predictably; this is a basis for the fluid model of Misra, Gong and Towsley [1] and for a growing literature on fluid models of congestion control. In this paper we argue that different fluid models arise from different buffersizing regimes. We consider the large buffer regime (buffer size is bandwidthdelay product), an intermediate regime (divide the large buffer size by the square root of the number of flows), and the small buffer regime (buffer size does not depend on number of flows). Our arguments use various techniques from queueing theory. We study the behaviour of these fluid models (on a single bottleneck link, for a collection of identical longlived flows). For what parameter regimes is the fluid model stable, and when it is unstable what is the size of oscillations and the impact on goodput? Our analysis uses an extension of the PoincaréLinstedt method to delaydifferential equations. We find that large buffers with droptail have much the same performance as intermediate buffers with either droptail or AQM; that large buffers with RED are better at least for window sizes less than 20 packets; and that small buffers with either droptail or AQM are best over a wide range of window sizes, though the buffer size must be chosen carefully. This suggests that buffer sizes should be much much smaller than is currently recommended.