Golestani, S.J. and S. Bhattacharyya. (1998). A class of end-to-end congestion control algorithms for the Internet. Proceedings of Sixth International Conference on Network Protocols.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....framework supporting the stability and performance properties of the system, while [11] which is based on control theory, makes different assumptions and proposes a different control technique. Two different approaches, also based on explicit congestion notification, are presented in [12, 13]. The second one is based on global network cost optimization; it operates through a distributed algorithm that converges to the optimal operating point if a well formed cost function is given. This procedure completely disregards transient behavior; this fact arises some concern about its ....

....based on a differential equation model. Charge sensitive TCP, with milestone works as [17, 18] is also often cited as a way to improve TCP performance. These schemes aims at finding a stable operation point for the network, disregarding the transient behavior (similarly to the work presented in [13]) during which the schemes cannot guarantee a low loss rate; however, Internet traffic is made of short connections, hence a scheme that controls the sources from the very beginning of the transmission is needed. In [19] we presented a specific implementation of the scheme discussed here that is ....

J. Golestani, S. Bhattacharyya. A Class of End-to-End Congestion Control Algorithms for the Internet. In IEEE ICNP'98, Oct. 1998.

....(minimization) of a utility (cost) function, subject to network resource constraints, where the constraints represent the interaction between different types of tra#c; i.e. tra#c with di#erent ingress egress nodes and or di#erent service requirements. In the paper by Golestani, et al. [7], instead of using link resource constraints, a link congestion cost is incorporated into the overall utility function. The optimization problem was then solved using a gradient type algorithm. Iterative algorithms were proposed where individual sources periodically adjust their sending rates ....

S. Golestani and S. Bhattacharyya, "A class of end-to-end congestion control algorithms for the Internet," in Proceedings of ICNP, October 1998.

....manage their own budgets avoiding waste. Several algorithms inspired on operational research and economy theories have been proposed to control resource usage in networks [7] These algorithms are able to converge towards a globally optimal resource allocation in a decen tralised way. In [9, 14, 15] such theories are applied to the problem of end to end congestion control, i.e. where bandwidth is the main scarce resource. In [9] an optimisation framework is used to derive a class of optimal algorithms inside which TCP, after some modification, can be seen as a special case. In [14] a ....

....to control resource usage in networks [7] These algorithms are able to converge towards a globally optimal resource allocation in a decen tralised way. In [9, 14, 15] such theories are applied to the problem of end to end congestion control, i.e. where bandwidth is the main scarce resource. In [9] an optimisation framework is used to derive a class of optimal algorithms inside which TCP, after some modification, can be seen as a special case. In [14] a thorough stability and fairness analysis of some optimisation based rate control algorithms is presented, and it is shown that these ....

S.J. Golestani, S. Bhattacharyya, "A Class of End-to-End Congestion Control Algorithms for the Internet", Proc. ICNP, October 1998.

....Several adaptive algorithms inspired on operational research and economy theories have been proposed to control resource usage in networks. These algorithms are able to converge towards a globally optimal resource allocation in a decentralised way. For an overview on the topic see [16] In [17, 18, 19] such theories are applied to the problem of end to end congestion control (i.e. where bandwidth is the scarce resource) In [17] an optimisation framework is used to derive a class of optimal algorithms inside which TCP (after some modification) can be seen as a special case. In [18] a thorough ....

....networks. These algorithms are able to converge towards a globally optimal resource allocation in a decentralised way. For an overview on the topic see [16] In [17, 18, 19] such theories are applied to the problem of end to end congestion control (i.e. where bandwidth is the scarce resource) In [17] an optimisation framework is used to derive a class of optimal algorithms inside which TCP (after some modification) can be seen as a special case. In [18] a thorough stability and fairness analysis of some optimisation based rate control algorithms is presented, and it is shown that these ....

S.J. Golestani, S. Bhattacharyya, "A Class of End-toEnd Congestion Control Algorithms for the Internet", Proceedings of ICNP, October 1998.

.... a number of enhancements have been proposed to TCP congestion control, including the persistent fast recovery of Newreno [14] selective acknowledgments (SACK) 22] and forward acknowledgments (FACK) 21] RFC 2581 describes the recommended algorithms for proper congestion control in TCP [1] [13] has formulated congestion control as a global optimization problem and has proposed a class of congestion control policies based on rewards and costs. While these TCP enhancements are interesting, significant recent trends in Internet applications and traffichave led to a renewed interest in in ....

GOLESTANI, S. J., AND S., B. A Class of End-to-End Congestion Control Algorithms for the Interney. In Proc. ICNP (1998).

....defined by the rate control algorithm. The optimum is characterized by a proportional fairness criterion. The work reviewed in Section 2 forms part of a growing body of research on the global optimization of networks, with important related approaches described by Golestani and Bhattacharyya [5] and Low and Lapsley [16] In Section 4 we describe a dynamical system which represents more closely the TCP algorithm, in particular the generalization MulTCP proposed by Crowcroft and Oechslin [2] and discuss its stability and fairness. Several authors have described network models of TCP ....

....in particular the generalization MulTCP proposed by Crowcroft and Oechslin [2] and discuss its stability and fairness. Several authors have described network models of TCP based on systems of differential equations, and the results of Section 4 are close variants on the work reported in [5], 6] 7] 13] 14] and [15] The dynamical system representations of Sections 3 and 4 model only gross characteristics of the flows through a network. And yet these flows evolve as a consequence of the fine detail of software operating at the packet level. In Sections 4 and 5 we briefly ....

S.J. Golestani and S. Bhattacharyya (1998) A class of end-to-end congestion control algorithms for the Internet. In Proc. Sixth International Conference on Network Protocols. www.bell-labs.com/user/golestani/

....Schemes like RAP and LDA can use a binomial algorithm (e.g. IIAD or SQRT) to avoid drastic rate reductions on encountering congestion. Golestani has formulated congestion control as a global optimization problem and has proposed a class of congestion control policies based on rewards and costs [31]. To combat the ill effects of multiplicative decrease on a single packet loss, various researchers have been looking at the class of equation based control algorithms [12] 32] 33] These are schemes where the sender measures the packet loss rate and round trip time over some past time and ....

S. J. Golestani and S. Bhattacharyya, "A Class of End-to-End Congestion Control Algorithms for the Internet," in Proc. ICNP, 1998.

....eciently manage their own budgets avoiding waste. Several algorithms inspired on operational research and economy theories have been proposed to control resource usage in networks [7] These algorithms are able to converge towards a globally optimal resource allocation in a decentralised way. In [9, 14, 15] such theories are applied to the problem of end to end congestion control, i.e. where bandwidth is the main scarce resource. In [9] an optimisation framework is used to derive a class of optimal algorithms inside which TCP, after some modi cation, can be seen as a special case. In [14] a thorough ....

....to control resource usage in networks [7] These algorithms are able to converge towards a globally optimal resource allocation in a decentralised way. In [9, 14, 15] such theories are applied to the problem of end to end congestion control, i.e. where bandwidth is the main scarce resource. In [9] an optimisation framework is used to derive a class of optimal algorithms inside which TCP, after some modi cation, can be seen as a special case. In [14] a thorough stability and fairness analysis of some optimisation based rate control 181 An Active Layered Multicast Adaptation Protocol ....

S.J. Golestani, S. Bhattacharyya, \A Class of End-to-End Congestion Control Algorithms for the Internet", Proc. ICNP, October 1998.

....Schemes like RAP and LDA can use a binomial algorithm (e.g. IIAD or SQRT) to avoid drastic rate reductions on encountering congestion. Golestani has formulated congestion control as a global optimization problem and has proposed a class of congestion control policies based on rewards and costs [31]. To combat the ill effects of multiplicative decrease on a single packet loss, various researchers have been looking at the class of equation based control algorithms [12] 32] 33] These are schemes where the sender measures the packet loss rate and round trip time over some past time and ....

S. J. Golestani and S. Bhattacharyya, "A Class of End-to-End Congestion Control Algorithms for the Internet," in Proc. ICNP, 1998.

....Congestion Control Consider a multicast group M. Let w j and j denote, respectively, the window size and the average round trip time, for a hypothetical unicast communication between the source and a receiver j 2 M . Let w min 4 = min j w j , and max 4 = max j j . As discussed in [3, 4], the multicast rate that can be attained by the window based approaches is r s = w min max ; On the other hand, let the rate of the bottleneck receiver be denoted as r min 4 = min j r j = w b = b , where b is the bottleneck receiver. Then, as shown in [3, 4] r s w min max w b b ....

....4 = max j j . As discussed in [3, 4] the multicast rate that can be attained by the window based approaches is r s = w min max ; On the other hand, let the rate of the bottleneck receiver be denoted as r min 4 = min j r j = w b = b , where b is the bottleneck receiver. Then, as shown in [3, 4], r s w min max w b b = r min : The above inequality states that the multicast rate attainable under the window based approaches is less than the rate of the bottleneck receiver. Since the rate attainable under the rate based approaches is only constrained by the rate of the bottleneck ....

S. J. Golestani and S. Bhattacharyya. A class of end-to-end congestion control algorithms for the Internet. Proc.of IEEE ICNP'98, Austin, TX, October 1998.

....control mechanism designed should be TCPfriendly, scalable, requires as little router support as possible, and judiciously detects network congestion and responds accordingly, based on (perhaps aggregated) acknowledgments from receivers. The most notable work in this category is reported in [81, 82, 83, 84, 85, 86, 87, 88]. Although not directly related to routing, these two issues should be considered, with multicast routing, as an integral part of a QoS centric multicast framework. It is also interesting to look into multicast routing, traffic engineering, and MPLS to provide, for example, differentiated ....

S. J. Golestani and S. Bhattacharyya. A class of endto -end congestion control algorithms for the internet. In Proc.of IEEE ICNP'98, October 1998.

.... a number of enhancements have been proposed to TCP congestion control, including the persistent fast recovery of Newreno [14] selective acknowledgments (SACK) 22] and forward acknowledgments (FACK) 21] RFC 2581 describes the recommended algorithms for proper congestion control in TCP [1] [13] has formulated congestion control as a global optimization problem and has proposed a class of congestion control policies based on rewards and costs. While these TCP enhancements are interesting, significant recent trends in Internet applications and traffic have led to a renewed interest in in ....

GOLESTANI, S. J., AND S., B. A Class of End-to-End Congestion Control Algorithms for the Interney. In Proc. ICNP (1998).

....of as components of a distributed computation to solve the global optimization problem. Thus, improvements in congestion control can proceed in a principled fashion, driven by improvements in the underlying optimization algorithm. While the optimization based approach has received much attention [13, 14, 15, 16, 17, 18, 19, 20, 21], it has not yet, to our knowledge, been applied to multicast congestion control. We will see in this paper, that applying this model to multicast o ers a formal foundation for developing fair multicast congestion control algorithms. The rest of this paper is structured as follows: In Section 2, ....

....to nd a distributed algorithm for solving (1 2) in which each individual session need only compute a local optimization to set its own rates. There is a growing body of research devoted to nding such a distributed algorithm and using it as a basis for unicast rate control in practical protocols [13, 14, 15, 16, 17, 18, 20, 19, 21]. Kelly [13] decomposes the basic optimization problem (1 2) into session and link subproblems, enabling the distributed computation of optimal rates. In addition to demonstrating the possibility of such a decomposition, Kelly also shows that the fairness properties satis ed by the resulting ....

S.J. Golestani and S. Bhattacharyya, \A class of end-to-end congestion control algorithms for the internet," in Proc. ICNP'98, 1998.

....computation to solve the global optimization problem. Thus, improvements in congestion control can proceed in a principled fashion, driven by improvements in the underlying optimization algorithm. While the optimization based approach has received much attention [2] 3] 4] 5] 6] 7] 8] [9], 10] it has not yet, to our knowledge, been applied to multicast congestion control. We will see in this paper, that applying this model to multicast offers a formal foundation for developing fair multicast congestion control algorithms. The rest of this paper is structured as follows: In ....

....for solving (1 2) in which each individual session need only compute a local optimization to set its own rates. There is a growing body of research devoted to finding such a distributed algorithm and using it as a basis for unicast rate control in practical protocols [2] 3] 4] 5] 6] 7] [9], 8] 10] Kelly [2] decomposes the basic optimization problem (1 2) into session and link subproblems, enabling the distributed computation of optimal rates. In addition to demonstrating the possibility of such a decomposition, Kelly also shows that the fairness properties satisfied by the ....

S.J. Golestani and S. Bhattacharyya, "A class of end-to-end congestion control algorithms for the internet," in Proc. ICNP'98, 1998.

....defined by the rate control algorithm. The optimum is characterized by a proportional fairness criterion. The work reviewed in Section 2 forms part of a growing body of research on the global optimization of networks, with important related approaches described by Golestani and Bhattacharyya [5] and Low and Lapsley [16] In Section 4 we describe a dynamical system which represents more closely the TCP algorithm, in particular the generalization MulTCP proposed by Crowcroft and Oechslin [2] and discuss its stability and fairness. Several authors have described network models of TCP ....

....in particular the generalization MulTCP proposed by Crowcroft and Oechslin [2] and discuss its stability and fairness. Several authors have described network models of TCP based on systems of differential equations, and the results of Section 4 are close variants on the work reported in [5], 6] 7] 13] 14] and [15] The dynamical system representations of Sections 3 and 4 model only gross characteristics of the flows through a network. And yet these flows evolve as a consequence of the fine detail of software operating at the packet level. In Sections 4 and 5 we briefly ....

S.J. Golestani and S. Bhattacharyya (1998) A class of end-to-end congestion control algorithms for the Internet. In Proc. Sixth International Conference on Network Protocols. www.bell-labs.com/user/golestani/

....with Section 7. 2 Background and Overview 2.1 Why Rate Based Congestion Control Consider a multicast group M. Let w j and j denote, respectively, the window size and the average round trip time, for a hypothetical unicast communication between the source and a receiver j 2 M . As discussed in [3, 4], the multicast rate that can be attained by the window based approaches is r s = w min max ; where w min = min j w j , and max = max j j . Let the rate of the bottleneck receiver be denoted as r min = min j r j = w b = b , where b is the bottleneck receiver. Then, we have r s w ....

....approaches is less than the rate of the bottleneck receiver. Since the rate attainable under the rate based approaches is only constrained by the rate of the bottleneck receiver, rate based congestion control is more effective for multicasts in terms of throughput. Moreover, as discussed in [3, 4], with the rate based approaches, throughput of a multicast session can be controlled to be independent of the longest round trip time among group members. To realize rate based congestion control, every member sends to the sender the maximum rate at which it can receive packets and the source is ....

S. J. Golestani and S. Bhattacharyya. A class of end-to-end congestion control algorithms for the Internet. Proc.of IEEE ICNP'98, Austin, TX, October 1998.

No context found.

S.J. Golestani and S. Bhattacharyya, "A Class of End-to-End Congestion Control Algorithms for the Internet," in Proc. ICNP, 1998.

....Scales of Congestion Control Although rate and window based congestion control have been studied for a long time, the real distinction between them is not always well understood. Here, we briefly summarize our view of these two approaches to congestion control; for a more complete discussion see [GB98] Let a unicast session s be subject to window based regulation , with the window size w s . Assuming a source that always has data to send, the amount of outstanding data will always 1 be w s . Denoting the average transmission rate by r s , and the average round trip time by s , it follows ....

....rate or window adjustments reflect random and instantaneous network events that are likely to pass even before the adjustment takes effect. Therefore, we adopt the viewpoint that regulation algorithms should update the session rates or window sizes based on quasi static network observations [GB98] In the case of window based regulation, the quasi static form of congestion control thus provided is enhanced by the more dynamic reaction to congestion that is inherent to the window scheme. 3.2 Extension of Window Congestion Control to Multicast Communications Before exploring the multicast ....

[Article contains additional citation context not shown here]

S. J. Golestani and S. Bhattacharyya. A class of end-to-end congestion control algorithms for the Internet. In Proc. 6th International Conference on Network Protocols, pages 137--150, 1998. http://www.belllabs. com/user/golestani/opt cong.ps.

....This is the time scale on which congestion control algorithms would enable a source to gradually adapt its transmission rate in response to changing network congestion levels. This is also the time scale on which notions such as average transmission rate and fairness are meaningful [14]. An obvious way of realizing worst path fairness is to have a multicast source adjust its rate in response to to loss indications from only the worst receiver in a multicast group and ignore all other loss indications. This corresponds to an LIF that filters out loss indications from all but ....

S.J. Golestani and S. Bhattacharyya. A Class of End-to-End Congestion Control Algorithms for the Internet. In Proc. ICNP, 1998. 25

.... One way of resolving this shortcoming is to couple this technique with end to end congestion control schemes that handle congestion problems on a quasi static basis while providing the desired fairness and or priorities in the amount of services given to different users in the long run (e.g. [18]) Finally, we mention a few more possible extensions of this work. First, it is sometimes advantageous to incorporate some packet dropping to address other network issues, such as aging of packets due to errors, and blocking caused when certain network links are overloaded [11] Second, if it ....

S. J. Golestani and S. Bhattacharyya, "A Class of End-to-End Congestion Control Algorithms for the Internet," Proc. 6th International Conf. on Network Protocols, Oct. 1998.

....This is the time scale on which congestion control algorithms would enable a source to gradually adapt its transmission rate in response to changing network congestion levels. This is also the time scale on which notions such as average transmission rate and fairness are meaningful ([14]) We now present a set of goals guiding the design of the loss indication filters, based on the notion of worst path fairness. Before doing so, we introduce some terminology that will be used throughout this paper. Let us consider a multicast session M , with N receivers numbered 1 through N . ....

S.J. Golestani and S. Bhattacharyya. A Class of End-to-End Congestion Control Algorithms for the Internet. In Proc. ICNP, 1998.

....performance over a wide range of network congestion conditions. In addition to my doctoral research on multicast flow and congestion control, I have been involved in a number of collaborative efforts the design and simulation of a class of TCP like congestion control algorithms for the Internet [3], the prototype implementation and performance evaluation of RMTP, a reliable multicast transport protocol [4] and the design and implementation of a link layer protocol for an ATM wireless LAN [5] Future Research Interests and Summary In the short term, I plan to continue research on multicast ....

....of my doctoral research I have acquired the ability to identify and understand fundamental issues and establish formal properties. At the same time, I have learned to develop solutions that are motivated by real world considerations. I have had the opportunity to collaborate with many researchers [3, 4, 6, 5], and have realized the importance of such collaborations in facilitating exchange of ideas and broadening one s perspective on research. I believe that that my record of research and collaboration will make me a strong addition to the networking research community. Supratik Bhattacharyya page 6 ....

S.J. Golestani and S. Bhattacharyya. A Class of End-to-End Congestion Control Algorithms for the Internet. In Proc. ICNP, 1998.

No context found.

Golestani, S.J. and S. Bhattacharyya. (1998). A class of end-to-end congestion control algorithms for the Internet. Proceedings of Sixth International Conference on Network Protocols.

No context found.

S.J. Golestani and S. Bhattacharyya, "A class of end-to-end congestion control algorithms for the Internet," Proc. ICNP, pp. 137--150, 1998.

No context found.

Golestani, S., Bhattacharyya, S.: A class of end-to-end congestion control algorithms for the internet. In: Proc. ICNP'98. (1998)

First 50 documents

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