Neural network control for TCP network congestion
31
Citation
21
Reference
10
Related Paper
Citation Trend
Abstract:
Active queue management (AQM) has been widely used for congestion avoidance in transmission control protocol (TCP) networks. Although numerous AQM schemes have been proposed to regulate a queue size close to a reference level, most of them are incapable of adequately adapting to TCP network dynamics due to TCP's non-linearity and time-varying stochastic properties. To alleviate these problems, we introduce an AQM technique based on a dynamic neural network using the back-propagation (BP) algorithm. The dynamic neural network is designed to perform as a robust adaptive feedback controller for TCP dynamics after an adequate training period. We evaluate the performances of the proposed neural network AQM approach using simulation experiments. The proposed approach yields superior performance with faster transient time, larger throughput, and higher link utilization compared to two existing schemes: random early detection (RED) and proportional-integral (PI)-based AQM. The neural AQM outperformed PI control and RED, especially in transient state and TCP dynamics variation.Keywords:
Random early detection
Network Dynamics
CUBIC TCP
Transient (computer programming)
Current Internet traffic is carried mainly by the Transmission Control Protocol (TCP), and the performance of the Internet depends strongly on how well TCP works. Detecting and mitigating the congestion is one of the main tasks of this protocol, in fact, different TCP versions are defined by their congestion control algorithm. Also, Active Queue Management (AQM) algorithms share the same goal of congestion mitigation with TCP, in particular, the most effective congestion control occurs when AQM and TCP work together. This paper presents a brief survey and a cross-comparison of the latest and most important TCP and AQM variants, then provides an evaluation of a different kind of performance (multiple Round-Trip Time, long delay, different congestion level, etc.) on the ns-3 network simulator over various types of environments. In a shared bottleneck, the choice of the TCP-AQM couple to adopt is crucial; we show that the results are not univocal, the "golden couple" depends on the performance that we want to boost and on the environment that we have to deal with.
CUBIC TCP
H-TCP
TCP Westwood plus
Explicit Congestion Notification
HSTCP
Cite
Citations (26)
With the increasing number of media flows using UDP in IP networks there is a strong need to enable fairness between UDP and TCP traffic. The TCP traffic will suffer because of the unresponsiveness of UDP to congestion in the IP network. To enable responsiveness of UDP traffic, the TCP-friendly rate control (TFRC) protocol can be used. With the coexistence of TCP and UDP, where the rate adaptation of UDP is managed by the TFRC protocol, knowledge is needed about how active queue management (AQM) algorithms perform to optimize networks performance. This paper studies the behavior of several AQM algorithms when small packet UDP flows compete with TCP flows in a heterogeneous network. A simulation study is presented, showing the results of different AQM algorithms. It shows that the adaptive AQM (AAQM) is very promising with respect to responsiveness and robustness.
CUBIC TCP
H-TCP
Cite
Citations (5)
Cite
Citations (10)
Explicit congestion notification (ECN), when used with random early detection (RED) gateways, reduces packet losses and delays of transport control protocol (TCP) based applications. However, choosing the buffer size and optimum parameter values of RED buffers are still open research issues. In this paper, we first present a model to determine the optimal value of RED's maximum threshold to achieve zero packet loss at RED gateways. Secondly, as an application of our model, we propose a new TCP algorithm, called differentiation capable TCP (Diff-C-TCP) to improve the TCP performance over lossy satellite links. Since most of network congestion losses can be eliminated by our zero loss model, Diff-C-TCP assumes packet losses to be indicators of link corruption, and uses ECN to explicitly indicate network congestion. We have shown that our zero packet loss analytical model matches simulation results very well, and our proposed Diff-C-TCP algorithm significantly improve TCP throughput.
TCP Vegas
Random early detection
CUBIC TCP
H-TCP
Packet loss
TCP Westwood plus
Cite
Citations (4)
Application requirements for delay and low jitter has driven the development of Active Queue Management (AQM) is very fast. Random Early Detection (RED) as one of the AQM has grown so rapidly and become a reference for the development of other AQM variants. RED to be fast growing because of its simplicity and ease to modified its parameter. There have been many studies that discuss the development of RED, but very few have focused on finding w q value, the weights for the optimal packet drop probability. In this study we tried to offer a different approach to the search w q values using genetic algorithms. This is done to adapt the possible values w q dynamically according to the character of traffic.
Random early detection
Hamming distance
Cite
Citations (1)
Over the years congestion has been a major issue affecting the internet leading to an increase in packet loss and delay. Researchers have proposed different algorithms to address the issue of congestion from Drop Tail, Early Random Drop to Active Queue Management (AQM). Random Early Detection (RED) is the first Active Queue Management (AQM) technique that was developed to support transport-layer congestion and decrease the impacts of network congestion on the router buffer. The idea behind RED is to sense and detect incipient congestion early and notify connections of congestion either by dropping packets arriving or by reducing its sending rate. Although various other AQM techniques have been proposed by researchers, RED is still the most commonly used algorithm for congestion avoidance and researches is still ongoing to enhance the performance of RED. In this paper, we have developed an extension to RED to address the limitation of RED and the algorithm is then compared with RED under various network scenarios. The results of the evaluation shows that the new method has outperformed RED.
Random early detection
Packet loss
Explicit Congestion Notification
Cite
Citations (12)
The interaction between Transmission Control Protocol (TCP) and Active Queue Management (AQM) can be captured using dynamical models. The communication network is viewed as a discrete-time feedback control system, where TCP adjusts its window size based on packet loss detection during the previous Round Trip Time (RTT) interval. In this research, a discrete-time dynamical model is proposed for the interaction between TCP and Random Early Detection (RED), one of the most widely used AQM schemes in today's Internet. The model, constructed using an iterative map, captures the detailed dynamical behaviour of TCP/RED, including the slow start and fast retransmit phases, as well as the timeout event in TCP. Model performance for various RED parameters is evaluated using the ns-2 network simulator. The results are then compared with other existing models. Based on evaluations and comparisons of our results, we can conclude that the proposed model captures the dynamical behaviour of the TCP/RED system. In addition, the model provides reasonable predictions for system parameters such as RTT, packet sending rate, and packet drop rate. An extension of the proposed model is also presented, which is a dynamical model of TCP with Adaptive RED (ARED), a modified AQM algorithm based on RED.
Random early detection
CUBIC TCP
H-TCP
TCP Westwood plus
Time-out
Cite
Citations (1)
Internet is considered as the worldwide system of many computers and their networks which are linked together by wired or wireless media. Network is based on internet protocol suite which has a model known as Transmission Control Protocol/Internet Protocol (TCP/IP) and a set of protocols. Model consists of TCP and IP as major protocols at transport and network layer respectively. TCP deals with end to end transmission and reassembly of packets. It also focuses on reliable and orderly transmission of packets. There are many TCP variants which are used in communication, such as Reno, Binary Increase Congestion control (BIC) and Vegas. TCP poses many issues like end to end delay in communication, congestion, connection management between sender and receiver end points. Congestion being one of the important issues affects the Quality of Service (QoS) of network and reduces overall network performance because of buffering of data in queue. Different Active Queue Management (AQM) algorithms which handle congestion are Random Early Detection (RED), Proportional Integral controller Enhanced (PIE), Controlled Delay (CoDel). These algorithms efficiently drop the packets when the buffer gets full or when it is near to full based on packet drop signal or other mechanisms. The article discusses about the importance of the right choice of TCP and AQM variant, results have been obtained by practical setup using flent and netpef traffic generator. The proposed work considers the study of different TCP and AQM variants for effective congestion control. Comparative study is conducted by considering selected TCP variant along with different AQM techniques. It reveals that TCP BIC along with CoDel provides less Iatency, high throughput and minimum packet drops.
Random early detection
CUBIC TCP
Explicit Congestion Notification
TCP Westwood plus
Compound TCP
TCP Vegas
Cite
Citations (5)
Active queue management (AQM) is an effective method used in Internet routers to enhance congestion control, and to achieve a trade off between link utilization and delay. The de facto standard, the random early detection (RED) AQM scheme, and most of its variants use queue length as a congestion indicator to trigger packet dropping. Random early detection (RED) is a widely studied active queue management (AQM) scheme implemented in routers for Internet congestion control. Although certain stability conditions for various RED controllers are discussed, no efficient method for dynamically adapting control gains or result on control gain optimization has been proposed so far. Therefore based on control theory, this paper proposes proportional and integral RED (PI-RED) algorithm, that not only considers the average queue length at the current time slot, but also takes into consideration the past average queue lengths within a round trip time. We provide guidelines for the selection of the feedback gains (the proportional parameter and the integral parameter) for TCP/RED system to stabilize the dynamics, make the congested queue converge at a certain target and hence to improve network performance. Simulations demonstrate that indeed satisfactory performance can be achieved if the control gains are selected based on the guidelines. Based on the stability condition and control gains selection method, extensive simulations with ns2 demonstrate that indeed satisfactory performance can be achieved in terms of robustness, drop probability and stability if the control gains are selected based on the guidelines
Random early detection
Robustness
Packet loss
Explicit Congestion Notification
Flow Control
Cite
Citations (1)
Random early detection
Robustness
Cite
Citations (3)