E. Altman, K.E. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses", Performance Evaluation, vol. 42, no. 2-3, pp. 129-147, Oct. 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Model of Packet Loss There has been extensive work on modeling packet loss. In [21] a small network is considered and a deterministic model for packet drops is developed. In [8] and [13] drops are assumed to be highly correlated over short time scales and independent over longer time scales. In [22], drops are assumed to be bursty. Furthermore, 22] makes no distinction between drop events and packet drops. Similarly, in this paper a drop event is when the congestion window divided by two, which can be made up of many lost packets, is referred to as a packet loss. In [12] drops events are ....

....on modeling packet loss. In [21] a small network is considered and a deterministic model for packet drops is developed. In [8] and [13] drops are assumed to be highly correlated over short time scales and independent over longer time scales. In [22] drops are assumed to be bursty. Furthermore, [22] makes no distinction between drop events and packet drops. Similarly, in this paper a drop event is when the congestion window divided by two, which can be made up of many lost packets, is referred to as a packet loss. In [12] drops events are modeled as a renewal process with various ....

E. Altman, K. Avrachenkov, and C. Barakat, "TCP in the presence of bursty losses," in ACM SIGMETRICS, June 2000.

....additional complexity into the behavior of network traffic. There has been a number of efforts aimed at gaining insights into this complex behavior, and by now the relationship between the throughput of a single TCP, its round trip time and loss probability is fairly well understood [2] 3] [4] [5] There remain, however, certain aspects of TCP that are not well understood and which cannot be analyzed readily with the models in these references. Among these outstanding issues we include the buffer behavior at a bottleneck router and the aggregate throughput that results from many TCP ....

E. Altman, K. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses," in Proceedings of SIGMETRICS, 2000.

.... pair that is subject to channel induced packet loss between TCP sender and receiver (e.g. 3] 4] 5] 6] On the other hand, the typical abstraction used for congestion loss is an equivalent per flow random loss model that randomly drops packets at the buffer (e.g. 7] 8] 9] 10] [11]) Note that while both abstract models are superficially the same, the causative loss mechanisms are quite different in nature. In the pioneering work [12] Lakshman and Madhow provide a complete analytical description of the (now well known) periodic TCP congestion window evolution over ideal ....

....(variable) queuing time by using the average measured round trip time. Thus, their model requires that both the average round trip time and the packet loss probability must be known. Other significant extensions to [8] also use the same assumptions of constant queuing delay (e. g [7] 9] 10] [11]) Congestion Loss Models (whether correlated or i.i.d) cannot be used for modeling the TCP dynamics over a wireless link since they assume: a) A known loss process at the packet layer (i.e. they do not attempt to derive the packet loss behavior from the underlying channel statistics) and (b) The ....

E. Altman, K. Avrachenkov, and C.Barakat, "TCP in presence of bursty losses," Computer Communication Review, vol. 30, no. 4, 2000.

....of introducing additional complexity into the behavior of network tra#c. There has been a number of studies in order to gain insights into this complex behavior, and by now the relationship between the throughput of a single TCP, its round trip and loss probability is fairly well understood [3, 14, 15, 17]. There remain, however, certain aspects of TCP that are not well understood and which cannot be analyzed readily with the models in these references. Among these outstanding issues we include the bu#er behavior at a bottleneck router and the aggregate throughput that results from many TCP flows ....

E. Altman, K. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses," in Proceedings of SIGMETRICS, 2000.

....additional complexity into the behavior of network traffic. There has been a number of studies in order to gain insights into this complex behavior. At the present, the relationship between the throughput of a single TCP and its round trip and loss probability is fairly well understood [3] 4] [6] [7] There are, however, certain aspects concerning TCP that are not well understood and which cannot be analyzed with the aforementioned models. This includes issues such as buffer behavior at a bottleneck router and the aggregate throughput when many TCP flows compete for the bandwidth of a ....

....chain and the loss probability (either independent or dependent on the state) determines the transition probabilities. Padhye et al. have derived an approximation for the steady state distribution of a Markov chain modeling a single TCP connection with a fixed loss probability [3] Altman et al. [6] have extended the model to cover more general loss processes, e.g. continuous time Markov chain with different loss probability in each state. 2.2 Extension of micro scale TCP model to many TCP flows In this section, we point to various efforts for extending the micro scale TCP model to many ....

[Article contains additional citation context not shown here]

E. Altman, K. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses," in Proceedings of SIGMETRICS, 2000.

....in the Internet. Most studies have been focused on a single TCP connection, using a variety of approaches: deterministic analysis of the steady state [9, 16] stochastic analysis of the steady state [20] fluid queueing model [7] algebraic computation [5] as well as some refined models of losses [17, 1, 2]. As it is well known, TCP reajusts its send rate through the feedback information on the network congestion. The interaction among TCP connections is thus a crucial and difficult issue. Gibbens and Kelly [11] analyzed the bandwidth sharing of different TCP connections using a fluid model where ....

Altman, E., Barakat, C. and Abratchenkov, K. "TCP in presence of bursty losses". Performance Evaluation, Vol. 42, Issues 2-3, pp. 129-147, 2000.

No context found.

E. Altman, K.E. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses", Performance Evaluation, vol. 42, no. 2-3, pp. 129-147, Oct. 2000.

No context found.

E. Altman, K. Avrachenkov, C. Barakat. TCP in presence of bursty losses. ACM SIGMETRICS (2000).

No context found.

E. Altman, K. Avrachenkov, C. Barakat. TCP in presence of bursty losses. ACM SIGMETRICS (2000).

No context found.

E. Altman, K. Avrachenkov and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun. 2000.

No context found.

E. Altman, K. Avrachenkov and C. Barakat, TCP in presence of bursty losses, Performance Evaluation, 42, (2000), 129--147.

No context found.

E. Altman, K. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun 2000.

No context found.

E. Altman, K.E. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses", Performance Evaluation, vol. 42, no. 2-3, pp. 129-147, Oct. 2000.

No context found.

E. Altman, K. Avrachenkov, C. Barakat, TCP in presence of bursty losses, ACM SIGMETRICS, Santa Clara, CA, 2000.

No context found.

E. Altman, K.E. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses", Performance Evaluation, vol. 42, no. 2-3, pp. 129-147, Oct. 2000.

No context found.

E. Altman, K. E. Avrachenkov, and C. Barakat. TCP in Presence of Bursty Losses. In Proceedings of ACM SIGMETRICS, 2000.

No context found.

E. Altman, K. Avrachenkov, C. Barakat, TCP in presence of bursty losses, in: Proceedings of ACM SIGMETRICS, June 2000.

No context found.

E. Altman, K. Avrachenkov and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun. 2000.

No context found.

Eitan Altman, Kostya Avrachenkov, and Chadi Barakat. TCP in Presence of Bursty Losses. In Performance Evaluation, volume 42, pages 129--147, October 2000.

No context found.

Eitan Altman, Kostya Avrachenkov, and Chadi Barakat. TCP in Presence of Bursty Losses. In Performance Evaluation, volume 42, pages 129--147, October 2000.

No context found.

E. Altman, K. Avratchenkov and C. Barakat, TCP in presence of bursty losses, in: ACM SIGMETRICS, 2000.

No context found.

E. Altman, K. Avrachenkov, and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun 2000.

No context found.

E. Altman, K. Avrachenkov and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun. 2000.

....from (2) X X X n 1 = Q n X X X n #D D D n . 3) We begin by studying the convergence and stability of the process X X X n . The next result follows from [8] or from Theorem 2A in [17] To show that the conditions of the theorem indeed hold, one may follow the approach in the appendix in [2]. E. Altman et al. Performance Evaluation 42 (2000) 129 147 133 Theorem 1. Assume that Y n contains a single recurrent class and is initially in steady state. Consider an arbitrary initial state X X X 0 . Then, X X X # n = # # # j=1 # # n 1 # i=n j Q i # #D D D n j 1 (4) is the ....

E. Altman, K. Avrachenkov, C. Barakat, TCP in presence of bursty losses, in: Proceedings of ACM SIGMETRICS, June 2000.

....this gure a more severe reduction due to multiple consecutive divisions of the congestion window by two. When a congestion appears, the network continues dropping packets over multiple round trip times resulting in these multiple consecutive divisions of the congestion window. In a previous paper [1], we presented a two state Markovian model to account for such a burstiness of losses. In that paper, we considered a lossy path with two states Good and Bad together with potential loss moments. The transmission rate may be reduced upon potential losses. A potential loss can transform into a real ....

....be reduced upon potential losses. A potential loss can transform into a real loss with probability pG in the Good state and with probability pB in the Bad state (pG pB ) The time between potential loss moments is assumed to be independently and identically distributed. Our main contribution in [1] was to show that the throughput of the ow control mechanism increases with the increase in burstiness of losses when keeping the average loss rate unchanged. We validated the model via simulations, but did not provide any algorithm for the identi cation of its parameters from real traces. We ....

[Article contains additional citation context not shown here]

E. Altman, K.E. Avrachenkov, and C. Barakat, \TCP in presence of bursty losses", Performance Evaluation, vol. 42, no. 2-3, pp. 129-147, Oct.

....sequel, we also refer to a batch of congestion signals as a loss event. We focus on the case when a certain limitation on the transmission rate exists. We determine the exact expression of the throughput under such a limitation. In the literature, only simplistic approximations have been proposed [3, 18] so far. We study two possible scenarios that lead to such a limitation: i) Peak Rate limitation: the limitation is not due to congestion in the network but rather to some external agreement. In that case, when the transmission rate reaches a certain level M , it remains constant until a loss ....

....is a special case of the model with a general stationary and ergodic arrival process studied in [4] For that model only bounds on the throughput were obtained. Exact expressions for the throughput were obtained there for the case in which no limitation on the transmission rate exists (see also [3, 13, 15, 18]) The paper is structured as follows. In Section 2 we describe a general model of flow control with limitation on the transmission rate and we provide a preliminary analysis. The two cases of peak rate limitation and congestion limitation are described separately in Sections 2.1 and 2.2. It is ....

[Article contains additional citation context not shown here]

E. Altman, K. Avrachenkov, C. Barakat. TCP in presence of bursty losses. ACM SIGMETRICS (2000).

....sequel, we also refer to a batch of congestion signals as a loss event. We focus on the case when a certain limitation on the transmission rate exists. We determine the exact expression of the throughput under such a limitation. In the literature, only simplistic approximations have been proposed [3], 19] so far. We study two possible scenarios that lead to such a limitation: i) Peak Rate limitation: the limitation is not due to congestion in the network but rather to some external agreement. In that case, when the transmission rate reaches a certain level , it remains constant until a ....

....is a special case of the model with a general stationary and ergodic arrival process studied in [4] For that model only bounds on the throughput were obtained. Exact expressions for the throughput were obtained there for the case in which no limitation on the transmission rate exists (see also [3], 14] 16] 19] The structure of the paper is as follows. In Section II we describe a general model of flow control with limitation on the transmission rate and we provide a preliminary analysis. The two cases of peak rate limitation and congestion limitation are described separately in ....

[Article contains additional citation context not shown here]

E. Altman, K. Avrachenkov and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun. 2000.

....sequel, we also refer to a batch of congestion signals as a loss event. We focus on the case when a certain limitation on the transmission rate exists. We determine the exact expression of the throughput under such a limitation. In the literature, only simplistic approximations have been proposed [3], 19] so far. We study two possible scenarios that lead to such a limitation: i) Peak Rate limitation: the limitation is not due to congestion in the network but rather to some external agreement. In that case, when the transmission rate reaches a certain level M , it remains constant until a ....

....is a special case of the model with a general stationary and ergodic arrival process studied in [4] For that model only bounds on the throughput were obtained. Exact expressions for the throughput were obtained there for the case in which no limitation on the transmission rate exists (see also [3], 14] 16] 19] The structure of the paper is as follows. In Section II we describe a general model of flow control with limitation on the transmission rate and we provide a preliminary analysis. The two cases of peak rate limitation and congestion limitation are described separately in ....

[Article contains additional citation context not shown here]

E. Altman, K. Avrachenkov and C. Barakat, "TCP in presence of bursty losses", ACM SIGMETRICS, Jun. 2000.

No context found.

E. Altman, K. Avrachenkov and C. Barakat, TCP in presence of bursty losses, Proceedings of SIGMETRICS 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