| S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report 2000-18, Dept. of Computer Sciences, Univ. of Texas, May 2000. |
....and therefore needlessly discriminated increases with the number of flows present in the system. A major drawback is the additional expense in implementation complexity. The properties of congestion feedback mechanisms based on AIMD and their fairness properties were discussed extensively [13] [16]. BAT by Bowen et al. 10] applies this background towards a rate controlled AQM scheme. This proves effective at keeping the queue length to a minimum while absorbing short term bursts. BAT approximates max min fairness [17] by managing the drop rates using an open loop control. Individual ....
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report 2000-18, Dept. of Computer Sciences, Univ. of Texas, May 2000.
....changes in network conditions Meanwhile, it is necessary to consider the convergence of congestion control schemes. Chiu and Jain [2] showed that AIMD control converges to fairness and efficiency. Recently, it was shown that the additive increase of TCP and general AIMD control is inferior [11]. Binomial algorithms [1] using non additive increase also possess the convergence property. Binomial algorithms are similar to AIMD in that they all use memory less control. That is, their control rules use only the current window size. Therefore, a question is, can one improve the convergence ....
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report TR
....along with three of the controls studied in [3] AIAD ( Additive Increase, Additive Decrease ) MIMD ( Multiplicative Increase, Multiplicative Decrease ) and the already mentioned superposition of such systems is stable [1] 4] This fact can be used to inductively prove global stability. 2 See [8] and [9] for examples. AIMD. MIAD ( Multiplicative Increase, Additive Decrease ) is missing because the other three controls suffice to explain the reasons that led to the choice of AIMD for TCP. Each axis in the diagram represents a user in the network. Therefore, any point (x1, x2) represents ....
....to the binary nature of the feedback, the system cannot converge to a stable point but to an equilibrium it will eventually fluctuate around the optimum. Note that these are by no means all possible controls: other examples are MIAD, controls with both an additive and multiplicative component [8] and nonlinear controls [9] 3. EXTENDED USE OF VECTOR DIAGRAMS In the design process of a congestion control mechanism, it is always important to distinguish between necessary and sufficient conditions: while the explanations in the last section suffice to prove that AIAD and MIMD are not ....
[Article contains additional citation context not shown here]
S. Gorinsky & H. Vin, Additive Increase Appears Inferior, Technical Report TR2000-18, Department of Computer Sciences, University of Texas at Austin.
....changes in network conditions Meanwhile, it is necessary to consider the convergence of congestion control schemes. Chiu and Jain [2] showed that AIMD control converges to fairness and efficiency. Recently, it was shown that the additive increase of TCP and general AIMD control is inferior [11]. Binomial algorithms [1] using non additive increase also possess the convergence property. Binomial algorithms are similar to AIMD in that they all use memory less control. That is, their control rules use only the current window size. Therefore, a question is, can one improve the convergence ....
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report TR
....are drastic changes in network conditions. Meanwhile, it is necessary to consider the convergence of congestion control schemes. Chiu and Jain [4] showed that AIMD control converges to fairness and eciency. Recently, it was shown that additive increase of TCP and general AIMD control is inferior [14]. Binomial algorithms [3] with nonlinear control (using non additive increase) also possess the convergence property. Binomial algorithms are similar to AIMD in that they all use memory less control. That is, their control rules use only the current window size. Therefore, a question is, can we ....
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report TR2000-18, Department of CS, Univ. of Texas at Austin, May 2000.
....increases the sending window until the packet loss caused by congestion occurs, where it rapidly reduce the sending window. With TCP Real, the sender can adjust the sending window before the packet loss occurs, thereby the fluctuation of the transmission rate is smaller. 1 A recent paper [7] discusses the inefficiencies of AIMD, although the issue is still a subject of discussion. 2 Source based congestion avoidance was first introduced in TCPVegas. The above modifications also constitute the foundation for an efficient recovery strategy over heterogeneous networks, which ....
S. Gorinsky and H. Vin, "Additive Increase Appears Inferior", Technical Report TR2000-18, Department of Computer Sciences, The University of Texas at Austin, May 2000.
....one packet per timeout. In this configuration, TCP would either be idle, or send at a rate higher than its fair share, due to its relatively coarse grain transmission rate. 1.3. 5 MAIMD In spite of the AIMD principle s popularity and public acceptance, it was recently disputed by the authors in [17]. In this work, there is an extensive analysis of AIMD in comparison with a different policy called MultiplicativeAdditive Increase Multiplicative Decrease (MAIMD) In MAIMD, the policy for increasing the sending rate (when the bandwidth utilization is under the desirable level) involves ....
....system matches the Internet, in that different flows are not synchronized in any sense, and the congestion control mechanism functions individually in each one of them. Applying MAIMD on an asynchronous system does not yield fairness convergence either. 7 Simulation experiments presented in [17] showed that TCP does not converge in terms of fairness in the case when one flow initiates a connection while another has already taken up all the available bandwidth. In the following years, it will become apparent if the perception of AIMD as presented in this paper will replace the common ....
S. Gorinsky and H. Vin, "Additive Increase Appears Inferior," tech. rep., University of Texas, Austin, 2000.
....TCP congestion control should favor AIMD over a stable LIMD algorithm with a multiplicative increase component Other criteria can be used to select an appropriate scheme for binary adjustments. For example, since LIMD provides superior convergence to ecient states in heterogeneous environments [4], TCP congestion avoidance mode can bene t from LIMD by acquiring promptly the bandwidth released by departed ows. Recall that TCP slow start uses multiplicative increase because additive increase converges to eciency slowly. Convergence to fairness under AIMD or LIMD (both rely on their ....
S. Gorinsky and H. Vin. Additive Increase Appears Inferior. Technical Report TR2000-18, Department of Computer Sciences, The University of Texas at Austin, May 2000.
No context found.
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report 2000-18, Dept. of Computer Sciences, Univ. of Texas, May 2000.
No context found.
Sergey Gorinsky and Harrick M. Vin. Additive increase appears inferior. Technical Report TR2000-18, Department of Computer Sciences, The University of Texas at Austin, May 2000.
No context found.
S. Gorinsky and H. Vin, "Additive increase appears inferior," Tech. Rep. TR2000-18, Department of Computer Sciences, The University of Texas at Austin, May 2000.
No context found.
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report 2000-18, Dept. of Computer Sciences, Univ. of Texas, May 2000.
No context found.
S. Gorinsky and H. Vin. Additive increase appears inferior. Technical Report 2000-18, Dept. of Computer Sciences, Univ. of Texas, May 2000.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC