J. Roberts, "Variable-Bit-Rate Traffic-Control in B-ISDN", IEEE Comm. Mag., pp. 50-56, Sept. 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....from different streams become synchronized, and produce clumps of cells; this produces an aggregate cell rate, in an interval of the order of the cell inter arrival time, larger than the link capacity. Cell loss at an ATM output link can arise from both cell scale and burst scale effects [Rob91] A burst scale effect takes place when the number of sources that are in their on phase, say aN , is such that their combined rate, which is viewed as a continuous fluid flow, exceeds the server capacity, i.e. aNh Nc; in this case, the buffer will start to fill. A cell scale effect takes ....

....probability less than 10 , when in fact the actual overflow probability is greater than 10 . Figure 3.2 shows the logarithm of the overflow probability plotted against the buffer size. Two distinct regimes are immediately evident, each dominated by cell scale and burst scale effects [Rob91] It is interesting to note that Figure 3.2 is similar to Figure 2.3 (page 31) which shows the logarithm of the overflow probability plotted against the buffer size for MPEG 1 traffic. Both figures show two distinct regimes, where different time scales are important for buffer overflow. In the ....

J. W. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Commun. Mag., pages 50--56, April 1991.

....= ph. Figure 1 shows the model for N = 4. Ton Toff 1 h 1 c 4 sources: Service times: Figure 1: Cell level model. Cell scale effects occur due to the synchronization of cell arrivals from different streams. Cell loss at an ATM output link can arise from both cell scale and burst scale effects [Rob91]. A burst scale effect takes place when the number of sources that are in their on phase, say aN , is such that their combined rate exceeds the server capacity, i.e. aNh Nc; in this case, the buffer will start to fill. A cell scale effect takes place when the synchronization of the sources ....

J. W. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Commun. Mag., pages 50--56, April 1991.

....the network provider the interesting issue is the video session s need for quality guarantees and the best way for providing it in the network [57] Capacity allocation could thus be deterministic or statistic. The choice of multiplexing mode for asynchronous transfers depends on several issues [86] [100] as illustrated in Fig. 15. Deterministic multiplexing is the natural choice when the peak rate is close to the link rate, when the mean is close to the peak rate, when traffic bursts are long compared to the buffers in the network, or when the quality requirements are strin gent. We may ....

J. W. Roberts, "Variable-Bit-Rate Traffic Control in B-ISDN," IEEE Communications Magazine, Vol. 29, No. 9, September 1991, pp. 50-56.

....start of an interval which is also fixed for the duration of the interval. The user benefit function and the price then determine howmany cells the user supplies during that interval. We are not concerned with the actual arrival distribution, as this is a cell scale component and can be buffered [10]. We concentrate on the burst scale component [10] and as long as the number of cells supplied in an interval is less than the total number that can be served in that interval we assume that there is no congestion. 3 PROBLEM FORMULATION We assume that the physical network is given. The problem ....

....duration of the interval. The user benefit function and the price then determine howmany cells the user supplies during that interval. We are not concerned with the actual arrival distribution, as this is a cell scale component and can be buffered [10] We concentrate on the burst scale component [10], and as long as the number of cells supplied in an interval is less than the total number that can be served in that interval we assume that there is no congestion. 3 PROBLEM FORMULATION We assume that the physical network is given. The problem we consider is how to induce the users to contend ....

J. W. Roberts, "Variable-bit-rate traffic control in B-ISDN", IEEE Commun. Mag., pp. 50-56, Sept. 1991.

....requested connection, either manually or in a menu driven environment. If users want to adjust connection parameters during a call then these parameter specifications may be automated. This kind of in call negotiation is desirable for bursty calls where the actual resource usage is of interest [20]. One traffic managementscheme for burst level resource reservation is described in [7] and a form of in call negotiation is now commerciallyavailable in some network interface equipment. Even for static connection requests, the call setup process mayinvolve a negotiation between the user and ....

J. W. Roberts, `Variable-Bit-Rate Traffic Control in B-ISDN', IEEE Commun. Mag.,Vol. 29, No. 9, pp 50-56, Sept. 1991.

....available to carry the connection at the quality of service (QoS) required by the user through the whole network, without altering the QoS of existing connections. One key feature of ATM is the statistical multiplexing of bursty connections which allows the network to achieve higher utilization (Roberts, 1991). Although many CAC schemes have been proposed in the literature (Fontaine, 1996b) none was deemed accurate and fast enough to be an adequate candidate to deal with variable bit rate (VBR) traffic. Most of the time, they provide solutions based on assumptions. These assumptions rely either on the ....

....system was trained only with the bufferless convolution algorithm (Fontaine, 1996a) but, it becomes obsolete when real traffic is used for training. A fourth input corresponding to the ratio between the buffer size and the mean burst size is considered to take into account the buffer s influence (Roberts, 1991) and (Boyer, 1995) Both approaches (3 and 4 input system) have been developed and trained with the same set of items coming from measurements. Each system has been trained with the backpropagation algorithm for 100 epochs (i.e. the training set has been used 100 times) 0.0 20.0 40.0 60.0 80.0 ....

Roberts, J.W. (1991) Variable-Bit-Rate traffic control in B-ISDN. IEEE Communications Magazine, pages 50--56.

....to determine, for a given traffic configuration and network utilization objective, the size of the buffer attached to the link so that the QoS objectives are met, especially with regard to cell loss. A huge amount of work has been devoted to this issue (see [16] for instance) Several studies [15], however, came to the conclusion that buffer dimensioning INRIA Asymptotic Results for the Superposition of a Large Number of Data Connections . 5 under burst scale congestion conditions is very uncertain because the performance of a queue in terms of cell loss and cell waiting time is then ....

J. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Communications Magazine, pages 50--56, September 1991.

....Due to the lack of consensus in this matter ITU T has approved in its recommendations only the peak cell rate parameter up to now [25] An accepted connection may suffer congestion leading to cell loss. In the literature two congestion phenomena (cell and burst time scale) are identified [38, 39, 40]. Here, we quote the definitions given by Roberts in [40] Cell time scale congestion is due to the simultaneous arrival of cells from independent sources and occurs even though the cell arrival rate is less than capacity. Burst time scale congestion happen when the cell arrival rate exceeds ....

.... only achieved either with small bit rate VBR sources (much small than link rate) or with larger buffer absorbing burst time scale delay [40] From a point of view of traffic control (resource allocation) it is worth distinguishing services according to how far they are adapted to sharing bandwidth [39]. 1 for example the network requires the sources slow down or stop transmission during congestion period 14 2.3.1 Best Effort Service In case of most of the LAN MAN and Internet applications with very bursty characteristics and of self similar nature [32] bandwidth reservation schemes seem to ....

J. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Communications Magazine, September 1991.

....the image quality varies from frame to frame. In VBR video coding algorithms, the size of the compressed frame varies from frame to frame and a constant image quality can be maintained [21, 26, 75] 3 VBR video coding algorithms are tailor made for packet switched networks for several reasons [36, 50, 58]. First, packet switched networks provide the flexibility of statistical resource sharing by several VBR video sources [39] Thus, if a video source does not use up all the bandwidth, another source can be multiplexed into the channel. CBR video algorithms, on the other hand, generally assume ....

J. Roberts, "Variable-bit-rate traffic control in B-ISDN", IEEE Communications Magazine, Vol. 29, No. 9, pp. 50-56, September 1991.

....imaging, computer aided design, and videotext services [18] By accessing image banks, these applications generate a significant amount of traffic. Depending on the resolution and coding techniques, each image is coded in a number of bits ranging from hundreds of kilobits to tens of megabits [26]. Image retrieval applications require moderate constraints on delay and are highly loss sensitive, compared to real time video. Voice traffic is delay sensitive but has a much lower bit rate than video traffic, as shown in Figure 2.2. The delay limit for voice traffic is considerably longer ....

J. Roberts, "Variable-Bit-Rate Traffic Control in B-ISDN," IEEE Communications Magazine, Vol. 29, No. 9, pp 50-56, Sept 1991.

....of a succession of images at irregular intervals. The traffic is of the on off type where traffic varies between activity periods at peak rate and silence intervals, depending on user database interactions. Display time compatible with user comfort requires a bit rate in the range of 1 to 10 Mb s [36]. The traffic has moderate constraints on the delay it can incur and is highly loss sensitive. Real time video, such as interpersonal video communication in an enhanced visual telephone service or in a video conferencing application, requires bit rates of up to several megabits per second. The ....

....with a single parameter. In this thesis, the definition of burstiness level as the ratio of the peak rate over the mean rate is adopted, because of its simplicity. 2. 4 Related Work on Traffic Shaping Although it is widely acknowledged that traffic shaping is essential for congestion control [18, 21, 32, 36, 38, 43], few researchers have actually investigated the effect of traffic shaping on the performance of the network. This section provides a brief review of some previous work that is relevant to this thesis. Sidi et al. 38] studied the smoothing effects of input rate regulation mechanisms (traffic ....

J. Roberts, "Variable-Bit-Rate Traffic Control in B-ISDN", IEEE Communications Magazine, Sept. 1991, pp. 50-56.

....m = ph. Figure 1 shows the model for N = 4. Ton Toff 1 h 1 c 4 sources: Service times: Figure 1: Cell level model. Cell scale effects occur due to the synchronization of cell arrivals from different streams. Cell loss at an ATM output link can arise from both cell scale and burst scale effects [Rob91]. A burst scale effect takes place when the number of sources that are in their on phase, say aN , is such that their combined rate exceeds the server capacity, i.e. aNh Nc; in this case, the buffer will start to fill. A cell scale effect takes place when the synchronization of the sources ....

J. W. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Commun. Mag., pages 50--56, April 1991.

....to determine, for a given traffic configuration and network utilization objective, the size of the buffer attached to the link so that the QoS objectives are met, especially with regard to cell loss. A huge amount of work has been devoted to this issue (see [16] for instance) Several studies [15], however, came to the conclusion that buffer dimensioning PI n1045 4 A. Dupuis, F. Guillemin and B. Sericola under burst scale congestion conditions is very uncertain because the performance of a queue in terms of cell loss and cell waiting time is then highly sensitive to the characteristics ....

J. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Communications Magazine, pages 50--56, September 1991.

....these elastic users. Traditionally the view has been that without an enforceable traffic description, no guarantees can be expected of the network. One possibility is that bursty sources can be well described during a burst. The need for burst level CAC has been widely discussed, e.g. 6] [7]; however burst start time and duration can be unpredictable. We take the view that adaptive users can be accommodated with a fast reservation scheme. We propose a form of in call negotiation on a timescale shorter than the dynamics of connection setup and teardown but longer than cell scale ....

J. W. Roberts, `Variable-Bit-Rate Traffic Control in B-ISDN', IEEE Commun. Mag., Vol. 29, No. 9, pp 50-56, Sept. 1991.

....and MMPP sources. The possible multiplexing degree of Bernoulli sources was studied in [23] and approximate results for the cell loss probability in an ATM multiplexer fed by ON OFF traffic are derived in [33] Some comments on controlling burst scale congestion and ACR are provided in [7] and [21]. In the following we present an exact analysis of the described model which requires a low computational effort. The analysis presented can be used to find proper choices of the source traffic descriptor parameter values for SCR enforcement. 3.2 Analysis using discrete time algorithm Again, we ....

J.W. Roberts, Variable-Bit-Rate Traffic Control in B-ISDN, IEEE Communication Magazine, September 1991, pp. 50-56.

....networks) i.e. packets must typically be fully received before they can be forwarded. As illustrated in Figure 1 (where W (t) stands for the amount of unfinished work in the system at time t) this can result in significant inaccuracies when estimating system performance (see also [52]) In [11] the authors propose and analyze models, that not only account for the type of dependencies captured by fluid flow models, but also preserve the discrete nature of arrivals and services that is characteristic of many communication systems. Before proceeding with the description of these ....

....it requires that the average length of the active and the idle periods on the link be proportional, i.e. within a factor g 1 = 1 Gamma g 1 ) This implies that for a given link utilization, the average duration of incoming bursts is fixed. Burst duration is, however, a key performance factor [30, 52], and it is of interest to develop models that allow burst duration and utilization to vary independently. In order to overcome this limitation, a model where the arrival process corresponds to an extended ON OFF process was presented and analyzed in [11] This model allows multiple packets with ....

J. W. Roberts, "Variable-Bit-Rate Traffic Control in B-ISDN", IEEE Commun. Mag., 29(9):50--56, September 1991.

....on the traffic descriptors of the requested connection and on the state of the network. ffl Bandwidth Allocation is an open issue related to the admission control end bandwidth enforcement. Multiple connections with variable bit rate traffic characteristics, may be statistically multiplexed [11] [9] [15] sharing the network resources, i.e. bandwidth, buffers, etc. The output channels capacity is less than the sum of the peak rates of all connections, but it exceeds their average total bandwidth requirements. Bandwidth allocation algorithms for statistically multiplexed channels should be ....

Roberts, W. James. "Variable-Bit-Rate Traffic Control in B-ISDN." IEEE Communications Magazine, pp. 50-56, Sept. 1991.

....m = ph. Figure 1 shows the model for N = 4. Ton Toff 1 h 1 c 4 sources: Service times: Figure 1: Cell level model. Cell scale effects occur due to the synchronization of cell arrivals from different streams. Cell loss at an ATM output link can arise from both cell scale and burst scale effects [Rob91]. A burst scale effect takes place when the number of sources that are in their on phase, say aN , is such that their combined rate exceeds the server capacity, i.e. aNh Nc; in this case, the buffer will start to fill. A cell scale effect takes place when the synchronization of the sources ....

J. W. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Commun. Mag., pages 50--56, April 1991.

....be an adequate assumption for communication systems (especially not in store and forward networks) i.e. packets must typically be fully received before they can be forwarded. As illustrated in Figure 1, this can result in significant inaccuracies when estimating system performance (see also [13]) It is one of the purposes of this paper to propose and to analyze models, that not only account for the type of dependencies captured by fluid flow models, but also preserve the discrete nature of arrivals and services that is characteristic of many communication systems. Before proceeding with ....

....= 1 Gamma g 1 ) This implies that for a given link utilization, the average duration of incoming bursts is fixed. Burst duration is, however, t Packet Arrivals t Packet Flow ON Period OFF Period Figure 3: System with ON OFF Source and Multiple Discrete Arrivals a key performance factor [13, 23], and it is of interest to develop models that allow burst duration and utilization to vary independently. This is the topic of Section 4. In Section 4 we consider a model where the arrival process corresponds to an extended ON OFF process. As in Section 3, we allow multiple packets with ....

J. W. Roberts. Variable-bit-rate traffic control in B-ISDN. IEEE Commun. Mag., 29(9):50--56, September 1991.

....available to carry the connection at the quality of service (QoS) required by the user through the whole network, without altering the QoS of existing connections. One key feature of ATM is the statistical multiplexing of bursty connections which allows the network to achieve higher utilization [3]. Although many CAC schemes have been proposed in the literature [4] none was deemed accurate and fast enough to be an adequate candidate to deal with variable bit rate (VBR) traffic. Most of the time, they provide solutions based on assumptions. These assumptions rely either on the assumed ....

....when the system was trained only with the bufferless convolution algorithm [9] but, it becomes obsolete when real traffic is used for training. A fourth input corresponding to the ratio between the buffer size and the mean burst size is considered to take into account the buffer s influence [3, 16]. Both approaches (3 and 4 input system) have been developed and trained with the same set of items coming from measurements. Each system has been trained with the backpropagation algorithm for 100 epochs (i.e. the training set has been used 100 times) 0.0 20.0 40.0 60.0 80.0 100.0 Epoch number ....

J.W. ROBERTS. "Variable-Bit-Rate traffic control in B-ISDN". IEEE Communications Magazine, pages 50--56, September 1991.

....the network provider the interesting issue is the video session s need for quality guarantees and the best way for providing it in the network [57] Capacity allocation could thus be deterministic or statistic. The choice of multiplexing mode for asynchronous transfers depends on several issues [86][100] as illustrated in Fig. 15. Deterministic multiplexing is the natural choice when the peak rate is close to the link rate, when the mean is close to the peak rate, when traffic bursts are long compared to the buffers in the network, or when the quality requirements are stringent. We may ....

J. W. Roberts, "Variable--Bit--Rate Traffic Control in B--ISDN," IEEE Communications Magazine, Vol. 29, No. 9, September 1991, pp. 50--56.

....when the sum of the peak rates is less than the output link capacity, i.e. N 1 f 1 N 2 f 2 N 3 f 3 f out , has been used to reduce the number of program runs. Furthermore, the peak rates f j should be less than 10 of the output link capacity in order to gain from statistical multiplexing [7 8]. 4.3.2 Generation of Sample Sets Consider the crucial task of selecting parameter combinations for cell loss evaluation. The goal is to determine a parameter distribution which enables correct ANN classifications. In this context it may be pointed out that the parameters f j , m j and t on(j) ....

Roberts J., Variable--bit--rate traffic control in B--ISDN, IEEE Communications Magazine, Vol 29, no. 9, pp 50--56, Sep 1991.

No context found.

J. Roberts, "Variable-Bit-Rate Traffic-Control in B-ISDN", IEEE Comm. Mag., pp. 50-56, Sept. 1991.

No context found.

J. W. Roberts, `Variable-Bit-Rate Traffic Control in B-ISDN', IEEE Commun. Mag.,Vol.29,No.9,pp 50-56, Sept. 1991.

No context found.

J.W. Roberts, "Variable Bit Rate Traffic control in B- ISDN," IEEE Communications Magazine, No. ( September 1991), pp 50-56

No context found.

J.W. Roberts. Variable-Bit-Rate Traffic Control in B-ISDN. IEEE Communications Magazine, 29(9):50--56, September 1991.

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