D. Mitra, Asymptotically optimal design of congestion control for high speed data networks, IEEE Transactions on Communications 40 (1992), 301-311.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Any such scheme can, therefore, at best be a congestionavoidance algorithm. Care must be taken to ensure that transmitters are not overly pessimistic. 8. 4 Congestion Control: Analysis and Simulation Several performance studies on congestion control algorithms have been reported re cently [2, 11, 26, 25, 30, 38, 40]. Several of them have reported that feedback delay plays an important role in stability and fairness. Most authors, however, have assumed a fixed (deterministic) delay because the actual distribution was not known. One contribution of this paper is an empirical observation on the distribution of ....

Mitra, D., "Asymptotically optimal design of congestion control for high speed data networks," IEEE Trans. Commun., pp. 301-311, February 1992.

....of a single file) and representing the cross traffic at each router by an exogenous flow, the model corresponds the open closed queueing network of Figure 2. RR n 3563 6 T. Bonald Source Destination Figure 2: Discrete model with cross traffic It turns out that this model, which was studied in [8, 9] in the particular case of exponential service times and Poisson arrival processes, is untractable in the more realistic situation where the service times are deterministic and the arrival processes generally distributed. In particular, it was shown in [2] that the throughput of the controlled ....

D. Mitra, Asymptotically optimal design of congestion control for high speed data networks, IEEE Transactions on Communications 40 (1992), 301-311.

....to be understood and used in studying resource management (e.g. in predictive controls, admission control, policing) Regarding control algorithms, networking algorithms that we are aware of do not really attempt to exploit the correlation structure in the underlying traffic. Example studies are [14, 23, 24, 25, 27, 31, 32]. The studies in [23, 32] develop a controller of the class discussed under sender initiated controls, assuming a known ARMA model. However, since traffic data is not used, no attempt is made to match models with measured traffic signatures. Our experience is that (i) accurate traffic models are a ....

Mitra, D., "Asymptotically optimal design of congestion control for high speed data networks," IEEE Trans. Commun., 40, (2), 301-311, February 1992.

....window to fill the network pipe (i.e. bandwidth delay product) Any further increase of the window size contributes only to increased queueing delay, not improved throughput. Optimal setting of the window size, however, requires knowledge of the RTT and the bandwidth delay product of the network [24]. Such information is usually not available at network elements. Instead, EWA determines when the network pipe is full by monitoring the occupancy of the buffer serving the outgoing ATM virtual circuit at the AAP: non empty buffer is either an indication of a full pipe or bursty traffic. At that ....

D. Mitra, "Asymptotically optimal design of congestion control for high speed data networks," IEEE Transactions on Communications, vol. 40, no. 2, pp. 301--311, February 1992.

....which are shared by several sources. Note that flow control is a term used for the end to end control of a single flow (or multiple flows on a per flow basis) Congestion control has been studied widely in computer networks and communications literature mainly for high speed data networks, [2, 3, 21, 24], and more recently for Asynchronous Transfer Mode (ATM) networks, 4, 10, 20, 25, 30] these references are only samples from a huge literature on the subject, they are by no means complete) There are two popular methods of controlling congestion. The rate based method is the one where ....

Mitra, D., "Asymptotically optimal design of congestion control for high speed data networks," IEEE Trans. on Communications, vol. 40 (1992), pp. 301--311.

....refinement of TCP congestion control algorithms. In this paper we revisit the fundamental question of the existence of fair end to end protocols and we provide a positive answer by constructing explicitly such protocols. 2 Model Window flow control is usually modeled as a closed queuing network [5, 6, 9]. For instance, in [5] the authors study the window flow control of a single connection with fixed propagation delay in a product form network. They derived the optimal window size and an adaptive window based control scheme based on the analytical model. In this paper we consider a closed ....

....control algorithms. In this paper we revisit the fundamental question of the existence of fair end to end protocols and we provide a positive answer by constructing explicitly such protocols. 2 Model Window flow control is usually modeled as a closed queuing network [5, 6, 9] For instance, in [5], the authors study the window flow control of a single connection with fixed propagation delay in a product form network. They derived the optimal window size and an adaptive window based control scheme based on the analytical model. In this paper we consider a closed multiclass fluid network ....

D.Mitra. Asymptotically optimal design of congestion control for high speed data network. IEEE Transaction on Communication, 40:301--311, February 1992.

....window to fill the network pipe (i.e. bandwidth delay product) Any further increase of the window size contributes only to increased queueing delay, not improved throughput. Optimal setting of the window size, however, requires knowledge of the RTT and the bandwidth delay product of the network [75]. Such information is usually not available at network elements. Instead, EWA determines when the network pipe is full by monitoring the occupancy of the buffer serving the outgoing ATM virtual circuit at the AAP: nonempty buffer is either an indication of a full pipe or bursty traffic. At that ....

D. Mitra, "Asymptotically optimal design of congestion control for high speed data networks," IEEE Transactions on Communications, vol. 40, no. 2, 301--311, February 1992.

....500 300 200 Graph 3 Goodput from above the transport layer for different desired throughputs, normalised to the VPC rate, for D1=25.5ms and D2=33.0ms. Similarly, increasing the end to end delay decreases the TCP performance, a phenomena that has already been documented in the literature [18]. This is shown in graph 4. Here, goodput vs desired load curves for different values of D1 and D2 were obtained, for a shaper size of 1000 cells. SMDS specifies that 95 of packets transferred through a DS 3 class access should experience a delay of less than 20ms between the two SNIs. However, ....

D.Mitra, "Asymptotically Optimal Design of Congestion Control for High Speed Data Networks", IEEE Transactions on Communications, Vol. 40, No.2, February 1992, pp301-311.

....of the flow control, in terms of utilization of the network resources. The main difficulty which arises in the analysis of current communication networks is that the traffic may exhibit periodicity [8] and long range dependence [16] which are not captured by traditional Markovian models [11, 13]. In addition, it turns out that in the particular case of window flow control, the throughput of the controlled connection depends in a crucial way on fine statistical characteristics of the cross flows, and not only on their traffic intensity. In [1] bounds on the performance of window flow ....

D. Mitra (1992) Asymptotically optimal design of congestion control for high speed data networks, IEEE Transactions on Communications, 40, 2, pp. 301-311.

.... 0 as p 0: Notice that E[ T errCost (N; p) in (1) is independent of the window size w: Figure 1 shows a few examples of the accuracy of the quasi independence assumption. Previous analytical studies have focussed on either flow control strategy or error control strategy in isolation (e.g. [2, 3, 5, 8, 6, 10]) The complexity of analyses has usually precluded a simultaneous study of both. Quasi independence asserts that sliding window flow control (with a fixed window size) and go back n error recovery are orthogonal when packet errors are independent. The result is of course provably valid only for ....

Mitra, D., "Asymptotically optimal design of congestion control for high speed data networks, " IEEE Trans. Commun., pp. 301-311, February 1992.

....is generally considered as a basis for future integrated telecommunications service. Since there would be an inevitable interaction and interference among users in the communication network, an increasing amount of research has been devoted to different control issues (see [1] 18] 19] 22] [23], 26] and their references) One of the basic problems arising here is the presence of propagation delays which pose a challenge for stability, since speed of data transmission in modern high speed networks keeps increasing. In most of the proposed algorithms and models (see [4] 5] 6] 11] ....

D.MITRA, Asymptotically optimal design of congestion control for high speed data networks, IEEETransactions on Communications, 40 (1992), pp. 301--311.

....scheme. Other analyses of flow control schemes include [22] 23] and [20] but these references do not address the specific concerns raised here in any detail. An alternative adaptive window flow control scheme that is of particular interest in the light of our findings is that proposed in [18], 19] The window adaptation mechanism there is designed to operate in a high bandwidthdelay product regime, and is based on asymptotics derived from a queueing model of the network. This scheme has the drawback of requiring more centralized coordination than TCP: the adaptation algorithm for ....

.... around an equilibrium value, rather than probing for bandwidth until there is a loss) than the drastic window size changes in TCP, so that a decentralized adaptive scheme based on a similarly smooth mechanism may overcome the drawbacks in TCP while not requiring the kind of knowledge assumed in [18], 19] Another possible mechanism is adaptation based on explicit feedback. Examples are the DECbit protocol [22] which suffers from some of the same problems as TCP) and the mechanisms recently proposed by the ATM Forum [1] for its ABR service class. Much further thought, however, is required ....

[Article contains additional citation context not shown here]

D. Mitra, "Asymptotically optimal design of congestion control for high speed data networks," IEEE Trans. Commun., vol. 40, no. 2, pp. 301-311, February 1992.

....window to fill the network pipe (i.e. bandwidth delay product) Any further increase of the window size contributes only to increased queueing delay, not improved throughput. Optimal setting of the window size, however, requires knowledge of the RTT and the bandwidth delay product of the network [24]. Such information is usually not available at network elements. Instead, EWA determines when the network pipe is full by monitoring the occupancy of the buffer serving the outgoing ATM virtual circuit at the AAP: non empty buffer is either an indication of a full pipe or bursty traffic. At that ....

D. Mitra, "Asymptotically optimal design of congestion control for high speed data networks," IEEE Transactions on Communications, vol. 40, no. 2, pp. 301--311, February 1992.

....large bandwidth delay product present in high speed networks, feedback methods such as the the adaptive window mechanisms proposed in [38] and [39] were considered infeasible. Recently, there has been a revision of this viewpoint, and results show that such schemes can be implemented ( 35] 36] [37]) The resurgence of interest in feedback based mechanisms is seen in works such as a threshold based feedback control scheme in the presence of non negligible propagation delays in [43] hop by hop flow control schemes that incorporate prediction of the states of other nodes have been shown to be ....

Mitra, Debasis, "Asymptotically Optimal Design of Congestion Control for High Speed Data Networks," IEEE Transactions on Communications, 40(2), pp. 301-311, 1992.

....and multicast services. Over the past several years, considerable effort has been directed at improving the existing techniques of congestion control in the Internet and at introducing new approaches to accommodate the requirements of new services and applications [RJ88, CJ89, Flo91, Kes91, Mit92, MS90, MS93, ZDE 93, Flo94, FJ93, FF96, Bra97] In this paper, we formulate the end to end control of user traffic in IP networks as a global optimization problem. The optimization framework enables us to bring out, in a comprehensive and concrete manner, the tradeoff between avoiding ....

....MCFC algorithm and some of the congestion control schemes previously proposed for the Internet. These schemes are the TCP congestion control [Jac88] currently used in the Internet, the Binary Feedback Scheme [RJ88] the Random Early Detection Gateways [FJ93] and the Dynamic Adaptive Windows [Mit92, MS90, MS93] While the global optimization framework is a foundation unique to the MCFC algorithm, there are important commonalities between the above schemes and the MCFC algorithm, regarding the underlying ideas or methods of execution. These common features allow us to apply some of the ....

[Article contains additional citation context not shown here]

D. Mitra. Asymptotically optimal design of congestion control for high speed data networks. IEEE Transactions on Communications, 40(2):301--311, February 1992.

....of data will be outstanding (sent but unacknowledged) at any given time. The test of a flow control protocol is its effectiveness in making network operation smoother as a result of this modification. The fundamental scaling dimension in the flow control problem is the bandwidth delay product [17, 18]. This is the amount of data that a source should keep outstanding in order to utilize network resources efficiently. When the bandwidth delay product (also referred to as the pipeline depth, or optimal window size; we use the symbol V) is small, then the flow control problem is not hard to solve. ....

D. Mitra, Asymptotically Optimal Design of Congestion Control for High Speed Data Networks, To Appear in IEEE Trans. on Communications, 1991.

....this oscillatory behavior does not fade. In [MUK91] it is proven that the oscillations are bounded, but no bound on their amplitude is explicitly found. Analogous situations are encountered in other increase decrease schemes in the presence of propagation delays; see for example [BOL90] In [MITR92], Mitra studies the effect of propagation delays in a high speed data network, under a sliding window flow control scheme. The approach is based on explicit modeling of propagation delays and an approximate asympotic analysis, where the delay bandwidth product is the large parameter. The ....

....product is the large parameter. The analytical results are used to determine the optimal window size and derive an adaptive window based control scheme for a single class multi hop network. Extensions to a multi class environment are presented in [MITR91] The network model used in [MITR92] is shown in Figure 9. This is the traditional multi hop virtual circuit model, where the round trip propagation delay model has been added. The controller queue is omitted due to the infinite data source assumption. 13 The round trip propagation delay is modeled as a infinite server system ....

[Article contains additional citation context not shown here]

Debasis Mitra, "Asymptotically Optimal Design of Congestion Control for High Speed Data Networks," IEEE Transactions on Communications, vol. 40, pp. 301--311, February 1992.

....choice of setpoint can be chosen without loss of generality, in Section 7, we show how modifying the setpoint can deal with the case where B is unknown. Work by Mitra et al. has shown that asymptotic analysis of product form queueing networks can be used to derive an optimal value of the setpoint [42, 43]. While their ideas are not directly applicable because of their assumptions of FCFS scheduling, Poisson cross traffic and an exponentially distributed packet service time distribution, to a first approximation, their results may be used to determine the choice of the optimal setpoint in the ....

.... 10 When control is done once every probe, it is easier to work in continuous time. We also make the fluid approximation [1] so packet boundaries are ignored, and the data flow is like that of a fluid in a hydraulic system. This approximation is commonly used [3, 6, 40, 55] and both analysis [43] and simulations show that the approximation is a close one, particularly when the bandwidth delay product is large. Let us assume that the input rate l is held fixed for some duration J. Then, n b (t J) n b (t) l(t) J (t) J (11) where is the average service rate in the time interval ....

[Article contains additional citation context not shown here]

D. Mitra, Asymptotically Optimal Design of Congestion Control for High Speed Data Networks, IEEE Trans. on Communications 40, 2 (Feb 1992), 301-311.

....the setpoint. In that case, a setpoint of B 2 balances the two tradeoffs. Of course, any other setpoint can be chosen with no loss of generality. Recent work by Mitra et al. has shown that asymptotic analysis of product form queueing networks can be used to derive an optimal value of the setpoint [18, 19]. The application of their ideas to this problem is explored in reference [20] 5.2. Frequency of Control We initially restrict control actions to only once per round trip time (RTT) this restriction is removed in section 9) For the purpose of exposition, divide time into epochs of length RTT ....

....below. If control is done once every probe, then it is easier to work in continuous time. We also make the fluid approximation [29] so packet boundaries are ignored, and the data flow is like that of a fluid in a hydraulic system. This approximation is commonly used [30, 31] and both analysis [19] and our simulations show that the approximation is a close one, particularly when the bandwidth delay product is large [20] Let us assume that l is held fixed for some duration J. Then, n b (t J) n b (t) l(t) J (t) J 11 where is the average service rate in the time interval [t, ....

D. Mitra, Asymptotically Optimal Design of Congestion Control for High Speed Data Networks, To Appear in IEEE Trans. on Communications, 1991.

No context found.

D. Mitra, "Asymptotically optimal design of congestion control for high speed data networks," IEEE Trans. Comm., 40 (2), 1992, pp. 301--311.

No context found.

D. Mitra, Asymptotically optimal design of congestion control for high speed data networks, IEEE Transactions on Communications 40 (1992), 301-311.

No context found.

D. Mitra (1992) Asymptotically optimal design of congestion control for high speed data networks, IEEE Transactions on Communications, 40, 2, pp. 301-311.

No context found.

D.Mitra, #Asymptotically optimal design of congestion control for high speed data network," IEEE Transaction on Communication 40, pp. 301#311, February 1992.

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