A. Charny, K.K. Ramakrishnan, and A. Lauck, Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks, IEEE/ACM Trans. on Networking, 4(4), August 1996, pp. 569-581.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....to increase his mean rate, since that would result in a decrease of his explicit rate. Such a feature of incentive compatibility is lacking in the existing flow control procedures where the network allocates in a fair way peak rates with no reference to the corresponding mean rates (e.g. see [CRL96, JKG 97, KVR95, and the references therein] and [ATMF96, Appendix I.5] 5.3.1 Controlling Effective Rates In this section we investigate implementation issues of the effective flow control procedure whose mathematical underpinnings are based on the pricing mechanism described by equations ....

A. Charny, K. K. Ramakrishnan, and A. Lauck. Time scale analysis and scalability issues for explicit rate allocation in ATM networks. IEEE/ACM Trans. on Networking, 4(4):569--580, August 1996.

.... queues serving the superposition of a large number of streams indicates circumstances when the busy period preceding a buffer overflow may be relatively short, and several authors have argued the advantages of preventing queue build up through the bounding of rates (see, for example, Charny et al. [13]) Any discussion of the performance of a rate control scheme must address the issue of fairness, since there exist situations where a given scheme might maximize network throughput, for example, while denying access to some users. The most commonly discussed fairness criterion is that of max min ....

Charny A, Ramakrishnan KK and Lauck A (1996). Time scale analysis and scalability issues for explicit rate allocation in ATM networks. IEEE/ACM Transactions on Networking , 4, 569--581. 31

....circuits, i.e. fixed paths which the cells associated with a given connection will follow. Thus ATM networks exhibit the same character as that of telephone networks. In designing the ATM standard, much attention was paid to efficiently managing bandwidth and providing QoS guarantees to users [12, 14, 29]. However, replacing current networks with new ATM net1 work infrastructure and protocols may take time and be costly. Another direction being pursued by researchers is to upgrade current Internet infrastructure and protocols (TCP IP) by increasing capacity and introducing service differentiation ....

....problem of achieving max min fairness. While there has been 4 much work in this area, we believe that many of the proposed mechanisms are not viable in a large scale networking environment where there are strong limitations on the complexity of the algorithms that can be implemented, see e.g. [14]. Our starting point is a simple mechanism for flow control proposed in [21] The rationale for the mechanism is as follows: suppose that n connections share a link with capacity c. If the capacity is to be shared evenly by the connections, then the fair rate e(t) n sesssions link capacity c ....

[Article contains additional citation context not shown here]

A. Charny, K. K. Ramakrishnan, and A. Lauck. Time scale analysis and scalability issues for explicit rate allocation in ATM networks. IEEE/ACM Trans. Networking, 4:569--581, 1996.

....feedback allows switches to specify a desirable traffic rate, so sources can rapidly adapt their traffic. Several algorithms have been suggested for computing the explicit rate. In general, the computation is based on the queue length, see e.g. 11, 17, 2] and or the arrival rates, see e.g. [48, 29, 10]. The former uses the difference between queue length and a target queue threshold to adjust the explicit rate. Algorithms using arrival rates to compute the explicit rate do so by dividing the capacity among sessions in a fair manner without considering the queue length. In order to divide ....

A. Charny, K. K. Ramakrishnan, and A. Lauck. Time scale analysis and scalability issue for explicit rate allocation in ATM networks. IEEE Trans. Networking, Vol. 4:569--581, 1996.

.... 13] Network configurations can be constructed which make algorithms of this class converge to unfair rate allocation [10] In the middle are clever implementations of the basic algorithm, which require significantly smaller amount of memory and enable efficient computation of the rate allocation [11]. All algorithms that recompute the allocation periodically are prone to problems that arise from the period being too short or too long. A long period makes an algorithm less responsive. A short period, when low speed flows are present, leads to errors in the measurement of traffic load on a ....

....[16] All switch algorithms for flow controlling the ABR ATM service have to deal with growth of switch queues during transients. Techniques for containing the queue growth include requiring source end stations to delay rate increases but to affect rate reductions immediately upon being notified [11, 12], setting aside a fraction of the link bandwidth [7 9, 13] etc. Both the techniques lower link bandwidth utilization. Delaying rate increases limits low utilization to periods of transients. But delaying rate increases requires mechanisms in the source endstation which have not been specified by ....

[Article contains additional citation context not shown here]

A.Charny et. al., "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks.," IEEE/ACM Trans. on Networking, pp 569 - 581, August 1996.

....8 3 units. Max min fairness is a well known concept [1] and weighted max min fairness is a fairly straightforward extension. In related work, there have been several distributed implementations for achieving max min fairness, but these have typically assumed per ow state in the network routers [23, 4]. In [23] core routers explicitly compute the fair share based on per ow state, but end hosts independently perform rate adaptation according 2 w(f1) 1 w(f2) 1 flow f1 link a B(a) 5 link b B(b) 5 flow f2 flow f4 w(f4) 4 flow f3 w(f3) 2 bg(f1) 1 bg(f2) 4 3 bg(f3) ....

A. Charny, k.K. Ramakrishnan, and A. Lauck. Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks. IEEE/ACM Transactions on Networking, vol. 4, no. 4, August 1996.

....fair, then a system optimum is achieved when users choices of charges and the network s choice of allocated rates are in equilibrium. We have not discussed convergence to equilibrium and an interesting and challenging question concerns whether rate control algorithms such as those described in [2], 3] and [4] may be adapted to implement the proportional fairness criterion described in this paper. A further challenging question concerns how the choice of parameter m s might be implemented in an ATM network. One possibility would be to use the Minimum Cell Rate of ATM standards [5] to buy ....

A. Charny, K.K. Ramakrishnan and A. Lauck, Time scale analysis and scalability issues for explicit rate allocation in ATM networks. IEEE/ACM Transactions on Networking 4, 569--581, 1996.

.... Transfer Mode (ATM) Forum on Trac Management has adopted rate based ow control as the prime mechanism for ow control of Available Bit Rate (ABR) trac in its networks (see, e.g. 5, 12, 30, 36, 38, 41] A widely accepted fairness criterion for rate based ow control is max min fairness [1, 3, 7, 8, 9, 19, 20, 21, 22, 24, 27], requiring that it be impossible for any session to receive a larger rate on the account of a session with a smaller or equal rate. Call max min rates those achieving max min fairness. Any rate based ow control algorithm can be classi ed into one of two broad classes, conservative and ....

.... has been scheduled for an increase, a control message loops around its path and calculates on its way the minimum share of bandwidth the session may take from the excess capacities of links along the path; this may require possible adjustments to the rates of con icting sessions (see, e.g. [7, 8, 9, 26, 27]) Roughly speaking, the convergence complexity of a rate based ow control algorithm is the number of rate adjustments performed in the worst case till all sessions are terminated and max min fairness is reached. Since reaching max min fairness fast is vital for ecient utilization of the network ....

A. Charny, K. K. Ramakrishnan, and A. Lauck, \Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks," IEEE/ACM Transactions on Networking, Vol. 4, No. 4, August 1996.

.... studied by Fatourou et al. 15, 16] A survey of work in the optimistic framework appears in [14] None of these previous works deals with priorities (with [21] being a single exception; see below) Asynchronous distributed algorithms that converge to max min fair rates have been presented in [7, 8, 9, 24, 31, 35] (this list is not exhaustive) all of them are conservative. Corresponding optimistic algorithms have been pointed out by Fatourou et al. 16] as a result of implementing in an asynchronous and distributed way corresponding algorithms presented in [15] which were described to work in a sequence ....

A. Charny, K. K. Ramakrishnan, and A. Lauck, "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks," IEEE/ACM Transactions on Networking, Vol. 4, No. 4, pp. 569--581, August 1996.

....the queue length, and thus cell loss, at a switch. It is especially crucial for multicasting services. We extend the definition of global feasibility to the multicasting environment. To ensure global feasibility, we extend the delayed increase policy proposed by Charny, Ramakrishan, and Lauck [5], to multicasting ABR sources. The delay period required to achieve global feasibility and the maximum cell loss in the absence of delayed increase policy are analyzed, both for switches using Siu and Tseng s algorithm, and for those using our new algorithm. 2 Related Work For multicasting flow ....

....value carried by this BRM cell; both have been carefully analyzed in Lemma B.1. 7 Global Feasibility Maintaining feasibility is an effective way to minimize the queue length, and thus cell loss due to buffer overflow, at a switch. Extending the definition given by Charny, Ramakrishan, and Lauck [5], the definition of feasibility for multicast flow, can be rewritten as follows, where SCCR s is the source current cell rate of session s, and NPATH sl is the number of paths of session s traversing link l. Definition 7.1 Global feasibility of transmission rate is defined as, for all links, at ....

[Article contains additional citation context not shown here]

A. Charny, K. K. Ramakrishnan, and A. Lauck, "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks", IEEE/ACM Trans. on Networking, Vol. 4, No. 4, pp. 569-581, Aug. 1996.

.... state [11] Distributed implementations of the max min algorithm have been proposed in literature, but they have long convergence times (O(3 Delta K Delta D) where D is the diameter of the network, and K is the number of distinct bottleneck links in the network) and feasibility constraints [12]. In a related work [25] we have implemented a maximumleaf spanning tree (MLST) based distributed approximation of the weighted max min algorithm which 4 In this paper, we assume that all flows are governed by the same revenue function. Revenue adjustments for different flows can be made in the ....

....(a) the revenue based rate adaptation algorithm proposed in Section 4, and (b) a max min fair allocation which is a special case of the weighted max min allocation where all flows have the same weight. Since the max min fair allocation has been proposed in literature for rate adaptation [12], we compare the two algorithms for a simple case when different flows traverse over different number of links. We establish three static flows: F1 along the path radha indra durga maruti, F2 along the path atri durga maruti, and F3 along the path atri durga. The Rspec for rate for all ....

[Article contains additional citation context not shown here]

A. Charny, K. K. Ramakrishna and A. Lauck, "Time scale analysis and scalability issues for explicit rate allocation in ATM networks," IEEE Transactions on Networking, Vol. 4, No. 4, August 1996.

....rate feedback allows switches to specify a desirable traffic rate, so sources can rapidly adapt their traffic. Several algorithms have been suggested for computing the explicit rate. In general, the computation is based on the queue length, see e.g. 7, 8, 2] and or the arrival rates, see e.g. [16, 12, 6]. The former uses the difference between queue length and a target queue threshold to adjust the explicit rate. Algorithms using arrival rates to compute the explicit rate do so by dividing the capacity among sessions in a fair manner without considering the queue length. In order to divide ....

A. Charny, K. K. Ramakrishnan, and A. Lauck. Time scale analysis and scalability issue for explicit rate allocation in ATM networks. IEEE Trans. Networking, Vol. 4:569--581, 1996.

.... S 5 A Control Tree Based Distributed Rate Adaptation Algorithm Most distributed rate adaptation algorithms based on max min fairness calculation proposed in the literature suffer from the following problems: a) very poor worst case and average case convergences, e.g. the algorithm proposed in [12] requires approximately six one way trips for convergence in the worst case scenario, b) feasibility problem [12] i.e. during the transience of adaptation, flows may experience severe transient packet loss. To maintain rate feasibility, the source rate increase policy is proposed in [12] ....

.... on max min fairness calculation proposed in the literature suffer from the following problems: a) very poor worst case and average case convergences, e.g. the algorithm proposed in [12] requires approximately six one way trips for convergence in the worst case scenario, b) feasibility problem [12], i.e. during the transience of adaptation, flows may experience severe transient packet loss. To maintain rate feasibility, the source rate increase policy is proposed in [12] decrease the source rate immediately if the adaptation requests the source for a rate decrease; defer the source 4 In ....

[Article contains additional citation context not shown here]

A. Charny, K. K. Ramakrishna and A. Lauck, "Time scale analysis and scalability issues for explicit rate allocation in ATM networks," IEEE Transactions on Networking, Vol. 4, No. 4, August 1996.

....the problem of achieving max min fairness. While there has been much work in this area, we believe that many of the proposed mechanisms are not viable in a large scale networking environment where there are strong limitations on the complexity of the algorithms that can be implemented, see e.g. [4]. Our starting point is a simple mechanism for flow control proposed in [5] The rationale for the mechanism is as follows: suppose that n connections share a link with capacity c. If the capacity is to be n sesssions link capacity c e(t) f(t) Figure 1: A network with one link and n sessions ....

.... send traffic at this explicit rate, the link flow (typically measured) will be n times e(t) i.e. f(t) ne(t) Now, since the number of active connections n may be unknown, we might estimate the number implicitly rather than monitoring it explicitly as other rate based control schemes do [4, 11]. One can estimate the number of active connections using n(t) f(t) e(t) The explicit rate is then computed based on the estimated number, i.e. e(t 1) c=n(t) Due to the capacity constraint, it is desirable to ensure that e(t) can not exceed the link capacity c, that is, e(t 1) min[ ....

[Article contains additional citation context not shown here]

A. Charny, K.K. Ramakrishnan, and A. Lauck. Time scale analysis and scalability issues for explicit rate allocation in ATM networks. IEEE/ACM Trans. Networking, 4:569--581, 1996.

....very well for maximizing the long term revenue. However, it is well known that max min and weighted maxmin algorithms require global state [8] Distributed implementations of the max min algorithm have been proposed in literature, but they have long convergence times and feasibility constraints [9]. In our testbed, we have implemented a maximum leaf spanning tree (MLST) based distributed approximation of the weighted max min algorithm which converges in O(log D) time in the average case and maintains feasibility at all times. While the details of this implementation are beyond the scope of ....

....(a) the revenue based rate adaptation algorithm proposed in Section 4, and (b) a max min fair allocation which is a special case of the weighted max min allocation where all flows have the same weight. Since the max min fair allocation has been proposed in literature for rate adaptation [9], we compare the two algorithms for a simple case when different flows traverse over different number of links. We establish three static flows: F1 along the path radha indra durga maruti, F2 along the path atri durga maruti, and F3 along the path atri durga. The Rspec for rate for ....

A. Charny, K. K. Ramakrishna and A. Lauck, "Time scale analysis and scalability issues for explicit rate allocation in ATM networks," IEEE Transactions on Networking, vol. 4, no. 4, August 1996.

.... TCP and TCP friendly congestion control [19, 15, 4, 3, 14] operate end to end using packet loss as a congestion indicator, and may require network element support for performance enhancement [16, 26] Our block differs from schemes like DECbit [32] TCP ECN [31] ATM ABR explicit rate control [9], hop by hop control [1, 24] credit based schemes [24] and others which may require varying degrees of computation, buffer management or fine grained scheduling support at bottlenecks. Our notion of edge to edge overlay is motivated by the diff serv model [5] and therefore is neither ....

....is increased based upon i (k Gamma 1) and the transmission is regulated by a token bucket, demand burstiness does not affect the network queues significantly. This also addresses the Use it or Lose it (UILI) problem of taking away the ingress rate allocation if demand does not exist to use it [9]. 11 6.2 Parameter Configuration Consider the parameters (rate increase interval) and T (measurement interval at ingress egress) The factors affecting these parameters are: maximum edge to edge system delay (excluding queuing delays) maximum changes in system delay, minimum transmission rate ....

Charny A. and Ramakrishnan K., "Time-scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks," IEE/ACM ToN, Vol. 4, No. 4, Aug '96.

....is that the iterative procedure requires scanning the state of each connection. Thus, it does not scale well with increasing number of connections. Scaling is a critical issue in ATM networks. The convergence and dynamic properties of this scheme as well as possible enhancements are discussed in [15]. Kalampoukas et al. 55, 56] proposed an explicit rate allocation algorithm that has similar convergence and performance properties with Charny s scheme but with significantly better scaling properties: on the arrival of an RM cell only the state of the connection it belongs to needs to be ....

A. Charny, K. K. Ramakrishnan, and A. Lauck, "Time scale analysis and scalability issues for explicit rate allocation in ATM networks," IEEE/ACM Transactions on Networking, vol. 4, no. 4, 569--581, 1996.

....for real time video: in the region of 100ms or so. Under this constraint, we want the requested rate to be smooth and not unnecessarily large, while at the same time we want the encoded rate to be close to or equal to the ideal rate. This also helps the rate allocation algorithms to stabilize [1]. When a given ideal frame size cannot meet the delay target given current buffer occupancy and the allocated rate, we encode at a lower rate so as to meet the constraint. We aim to request and be allocated sufficient bandwidth such that this happens rarely. The SAVE algorithm achieves this aim ....

....they have been cropped more than 20 . Robustness to Network Feedback Delay. The rate allocation mechanism of the explicit rate network is not expected to instantaneously allocate a rate in response to requests. To include the time needed for the stabilization of the rate allocation algorithms [1], we need to verify that quality measures are preserved even for a relatively large feedback delay (considered in number of frame times) Even in the absence of network congestion, frame cropping is likely occur when the short term average demand suddenly changes. Therefore, some frames are likely ....

A. Charny, K.K. Ramakrishnan, A. Lauck, "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks", IEEE/ACM Transactions on Networking, Vol. 4, No. 4, August 1996.

....for real time video: in the region of 100ms or so. Under this constraint, we want the requested rate to be smooth and not unnecessarily large, while at the same time we want the encoded rate to be close to or equal to the ideal rate. This also helps the rate allocation algorithms to stabilize [1]. When a given ideal frame size cannot meet the delay target given current buffer occupancy and the allocated rate, we encode at a lower rate so as to meet the constraint. We aim to request and be allocated sufficient bandwidth such that this happens rarely. The SAVE algorithm achieves this aim ....

....they have been cropped more than 20 . Robustness to Network Feedback Delay. The rate allocation mechanism of the explicit rate network is not expected to instantaneously allocate a rate in response to requests. To include the time needed for the stabilization of the rate allocation algorithms [1], we need to verify that quality measures are preserved even for a relatively large feedback delay (considered in number of frame times) Even in the absence of network congestion, frame cropping is likely to occur when the short term average demand suddenly changes. Therefore, some frames are ....

A. Charny, K.K. Ramakrishnan, A. Lauck, "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks", IEEE/ACM Transactions on Networking, Vol. 4, No. 4, August 1996.

....distributed explicit rate allocation algorithm should converge to the eventual max min fair allocations [2] for the individual flows within a reasonably short time. Simplistically, the time it takes to converge is approximately twice the round trip time multiplied by the number of bottleneck rates [3]. This convergence time is based on having a stable demand from the individual sources during the period of convergence of the distributed algorithm. Our smoothing algorithm attempts to keep the short term variability in the demand relatively small, thus helping the network s explicit rate ....

....demand will be helpful in achieving better average fairness compared to when the source rates change on a much shorter time scale. For a detailed description of the end system policies and switch policies that assure max min fairness while maintaining small queues, we refer the reader to [1] [3], 11] 15] We shall describe the operation of SAVE by means of the following data rates illustrated in Figure 1: ffl Ideal Rate: This is the rate that is required by the en 5 10 15 Frames Smoothed 5 10 15 Superposition Size 2 4 6 99.9 as multiple of mean 5 10 15 Frames Smoothed 5 10 ....

Charny, A., Ramakrishnan, K. K., Lauck, A., "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks", IEEE/ACM Transactions on Networking, Vol. 4, No. 4, August 1996.

.... distributed explicit rate allocation algorithm should converge to the eventual max min fair allocations for the individual flows within a reasonably short time [1] Simplistically, the time it takes to converge is approximately twice the round trip time multiplied by the number of bottleneck rates [2]. This convergence time is based on having a stable demand from the individual sources during the period of convergence of the distributed algorithm. Our smoothing algorithm attempts to keep the short term variability in the demand relatively small, thus, helping the network s explicit rate ....

....will be helpful in achieving better average fairness compared to when the source rates change on a much shorter time scale. For a detailed description of the end system policies and switch policies that assure max min fairness while maintaining small queues, we refer the reader to [14] 9] [2], 17] We define the following different rates for this environment: ffl Ideal Rate: This is the rate that is required by the encoder to code the frame at ideal quality. ffl Encoded Rate: This is the rate given to the encoder based on the algorithm for smoothing and rate adaptation. We assume ....

Charny, A., Ramakrishnan, K. K., Lauck, A., "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks", IEEE/ACM Transactions on Networking, Vol. 4, No. 4, August 1996.

....feasibility (i.e. the capacity of any resource in the network primarily link bandwidth is not exceeded at any time) With even a small excess in the source rate, it can cause a substantial queue buildup. The ABR service can ensure a particular notion of fairness max min fairness [4, 12], which requires a distributed computation [3] An incremental computation that scales with the number of connections is described in [9] The incremental computation of source rates can potentially result in the rates being infeasible for short intervals (often one round trip time) Varying ....

....queue buildups which are sometimes substantial. One way to ensure feasibility is to force a source being allowed to increase its rate to delay any increase until all other sources have received and implemented their decreases. Thus, the aggregate rate at a given link will never exceed its capacity [4]. This method introduces considerable delay to sources when they start up or when they are asked to increase their rate, thus impacting applications (and userperceived performance) adversely. It also may lead to underutilization of resources. In addition, when the bandwidth in the network ....

[Article contains additional citation context not shown here]

Charny A., Ramakrishnan, K.K., and Lauck, A., "Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks" IEEE/ACM Transactions on Networking, August 1996.

No context found.

A. Charny, K.K. Ramakrishnan, and A. Lauck, Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks, IEEE/ACM Trans. on Networking, 4(4), August 1996, pp. 569-581.

No context found.

A. Charny, K. K. Ramakrishnan, and A. Lauck, #Time Scale Analysis and Scalability Issues for Explicit Rate Allocation in ATM Networks," IEEE#ACM Trans. on Netwroking, vol. 4, no. 4, pp. 569#581, August 1996.

No context found.

A. Charny, K. K. Ramakrishnan, and A. Lauck, "Time scale analysis and scalability issue for explicit rate allocation in ATM networks," IEEE/ACM Trans. Networking, vol. 4, pp. 569--581, Aug. 1996.

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