S. Floyd, Comments on measurement-based admissions control for controlled-load services, LBNL, Tech. Rep., 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....dynamic estimation of available capacity based on the aforementioned notions and tools. Call admission control based on traffic measurements has been proposed by several researchers as an attractive solution in supporting real time services over a network offering statistical guarantees [7] 8] [9], 10] By using traffic measurements instead of traffic descriptors, the levels of utilization achieved are higher than with static CAC, without compromising QoS. In [11] and [9] Chernoff bounds are employed in the estimation of available capacity using ON OFF traffic approximation arguments. ....

....as an attractive solution in supporting real time services over a network offering statistical guarantees [7] 8] 9] 10] By using traffic measurements instead of traffic descriptors, the levels of utilization achieved are higher than with static CAC, without compromising QoS. In [11] and [9], Chernoff bounds are employed in the estimation of available capacity using ON OFF traffic approximation arguments. ON OFF Markov as well as Normal approximations of available capacity are given in [12] Also, in [13] measurement based CAC is performed by employing estimates of buffer overflow ....

[Article contains additional citation context not shown here]

S. Floyd, "Comments on measurement-based admissions control for controlled-load services," Tech. Rep., Lawrence Berkeley National Laboratory, July 1996.

....the path. While RSVP provides guaranteed quality of service (i.e. integrated service) it has significant scalability problems, and has not been widely deployed in today s Internet. Instead of providing such a hard performance guarantee, several measurement based admission control algorithms [11, 14, 6, 9] have been proposed to provide a soft guarantee. Allowing occasional performance degradation enables more efficient utilization of network resources while still providing acceptable service. There have been several recent proposals on endpoint admission control that use the differentiated service ....

S. Floyd. Comments on Measurement-based Admissions Control for Controlled-Load Services. submitted to Computer Communications Review, July 1996.

....flow specific delay violation probability guarantees. Several existing measurement based admission control (MBAC) algorithms address flow QoS requirements along the dimensions of the bandwidth or aggregate loss rate. Breslau et al. 4] performed a comparative study of several MBAC algorithms [17, 12, 8, 10, 5] under FIFO service discipline and concluded that none of them are capable of accurately achieving loss targets. Qiu and Knightly[17] proposed an MBAC scheme that measures maximal rate envelopes of aggregate traffic flows. Boorstyn et al. 3] developed the notion of effective envelopes that capture ....

S. Floyd. Comments on measurement-based admission control for controlled load services. Technical Report, Lawrence Berkeley Laboratory, July 1996.

....flows and are especially suited to provide flow specific delay violation probability guarantees. Several existing MBAC algorithms address flow QoS requirements along the dimensions of the bandwidth or aggregate loss rate. Breslau et al. 7] performed a comparative study of several MBAC algorithms [35, 23, 17, 19, 9] under FIFO service discipline and concluded that none of them are capable of accurately achieving loss targets. Qiu and Knightly[35] proposed an MBAC scheme that measures maximal rate envelopes of aggregate traffic flows. Boorstyn et al. 6] developed the notion of effective envelopes that capture ....

S. Floyd. Comments on measurement-based admission control for controlled load services. Technical Report, Lawrence Berkeley Laboratory, July 1996.

....buffer dynamics, even in the case of heterogeneous flows with highly bursty (temporally correlated) traffic. Moreover, as the envelope does not employ a central limit Gaussian traffic approximation, it can characterize scenarios with moderate statistical multiplexing gains: as detailed in [11], this is important in link sharing scenarios with a potentially moderate number of flows per traffic class. Next, exploiting properties of this measured aggregate traffic envelope, we devise a new envelope based MBAC algorithm as follows. First, subject to the (temporary) assumption that the ....

....time scale is a fundamental one to MBAC, and further discussions of its proper setting can be found in [4] 16] and [18] for example. III. SIMULATION EXPERIMENTS In this section, we evaluate the performance of the new algorithm for measurement based admission control and compare with [11], 18] The workload consists of a set of twenty 30 min traces of MPEG and JPEG compressed video from [26] In addition, we perform simulations using heavy tailed on off sources, which also form a long range dependent traffic flow in aggregate. With this collection of traces and an implementation ....

[Article contains additional citation context not shown here]

S. Floyd, "Comments on measurement-based admissions control for controlled-load services," Lawrence Berkeley Laboratory, Tech. Rep., July 1996.

....path computations and call admissions without any resource reservations. This helps in reducing the system complexity and achieving relatively fast QoS provisioning. It is well known that multimedia applications are likely to produce highly bursty tra#c with hardly predictable characteristics [8, 9]. This high burstiness keeps measurement based approaches from properly performing. This is because the severe tra#c fluctuation misleads the approaches into producing non feasible QoS paths and making incorrect call admission decisions. The system that we propose carries out rather reliable call ....

S. Floyd. Comments on Measurement-based Admissions Control for Controlledload Services, 1996.

....whereby traffic transported over UDP can also be made to behave fairly. Such schemes can be generally split into two: equation model based, and sender receiver based. Model based schemes use mathematical models of TCP to define behaviour that leads to a fair share resource allocation [Padhye99, Floyd00]. Sender receiver based schemes perform rate control as in TCP, utilising some system of acknowledgement for successfully received data [Sisalem98, Rejaie99, Rhee00] This enables the relevant end point to implement an additiveincrease, multiplicative decrease rate control in a similar manner to ....

....transmission rates in order that resources may be reserved at connection setup [ATMF TM99] 2.2. 3 Measurement based admission control An alternative to requiring that the connection explicitly declare its traffic parameters is to use MBAC (MEASUREMENT BASED ADMISSION CONTROL) Gibbens95, Floyd96, Gibbens97, Jamin97a, Jamin97b, Wang99] In this case, the network measures its current load and then uses these measurements to make a decision about whether it should accept a new connection. This approach has the advantage that it relaxes the requirement that users or applications know a ....

S. Floyd. Comments on Measurement-based Admission Control for Controlled-Load Services. Technical Report, Lawrence Berkeley National Laboratory, July

....if these values were as equally valid as traffic descriptors given a priori may, therefore, lead to a degrading of the MBAC algorithm s ability to honour the QoS to which it is committed. The ability to reuse static AC algorithms has meant many MBAC algorithms are based upon the CE model (e.g. Floyd96, Casetti96, Gibbens97, Jamin97c, Jamin97b, Jamin97c, Droz97] and [Lewis98] However, this does not mean the problem with measurement uncertainty has been left unconsidered. A common approach used to overcome the disadvantage introduced by this error is to compensate with a conservative ....

....Droz97] and [Lewis98] However, this does not mean the problem with measurement uncertainty has been left unconsidered. A common approach used to overcome the disadvantage introduced by this error is to compensate with a conservative measurement procedure; examples of this approach include [Floyd96, Gibbens97, Jamin97c] A slight variation is given in [Casetti96] an algorithm that adjusts its behaviour on the basis of the traffic offered. The MBAC algorithms from [Grossglauser97b, Tse99, Gibbens95, Key95, Courcoubetis95] and [Duffield99a] depart from the CE model. Each of these algorithms ....

[Article contains additional citation context not shown here]

S. Floyd. Comments on Measurement-based Admissions Control for Controlled-Load Services. Technical Report, Lawrence Berkeley Laboratory, July 1996. (pp 58, 59, 105, 107, 176, 177)

....measured sum approach, which bases admission decisions on an estimate of the mean aggregate arrival rate. Gibbens et al. 13, 14] study Chernoff bound based admission rules. They assume on off traffic and consider a tangent on the effective bandwidth function in their admission rule. Floyd [15] as well as Brichet and Simonian [16] study measurement based admission control based on the Hoeffding bound. They employ an exponential weighted moving average measurement mechanism. All these approaches have structural similarities, which are studied by Jamin and Shenker [17] Roughly speaking, ....

....is derived from the logarithmic moment generating function, but use the actual logarithmic moment generating function in our admission rule. The LD admission rule thus takes the mean as well as the higher moments of the measured aggregate arrivals into account. Hoeffding Bound Approach Floyd [15] studies measurement based admission control based on the Hoeffding bound [41, 42] The Hoeffding bound is a Chernoff style bound for sums of bounded, independent random variables. Recall that U j ; j = 1; J , are steady state random variables denoting the rate at which connection j ....

[Article contains additional citation context not shown here]

S. Floyd. Comments on measurement--based admission control for controlled--load services. submitted, July 1996.

....the level of service failures that will result. The other is to achieve the highest possible utilization for a given level of service failures. Several measurement based admission control algorithms have been proposed in the literature (see, for example, 17] 10] 14] 15] 16] 19] 20] [11], 7] 13] 21] and they implicitly or explicitly seek to achieve one or both of these design goals. The proposed algorithms, although embracing similar goals, differ in four important ways. First, some algorithms are principled, based on solid mathematical foundations such as Large Deviation ....

....if the sum of the token rate of the new flow and the estimated rate of existing flows is less than a utilization target times the link bandwidth. A time window estimator is used to derive the estimated rate of existing flows. Hoeffding Bounds (HB) The admission control algorithm described in [11] computes the equivalent bandwidth for a set of flows using the Hoeffding bounds. A new flow is admitted if the sum of the peak rate of the new flow and the measured equivalent bandwidth is less than the link utilization. An exponential averaging measurement mechanism is used to produce the load ....

[Article contains additional citation context not shown here]

FLOYD, S. Comments on measurement-based admissions control for controlled-load services. Technical report, Lawrence Berkeley Laboratory, July 1996.

....rather than relying on worst case analysis. Measurement based admission control algorithms (MBACs) have been shown to achieve much higher utilization than parameter based admission control algorithms [5] There has been much research interest in MBACs, and several new MBACs have been proposed; see [2, 3, 4, 6, 7] for some examples. This research has typically focused on the equations (and, to a lesser extent, the measurement algorithms) used to decide whether or not to accept an incoming flow. These proposals range from the simple and ad hoc to the complicated and principled, derived from detailed ....

Floyd, S. Comments on measurement-based admissions control for controlled-load services. Technical report, Lawrence Berkeley Laboratory, July 1996.

....for audio, video, and other so called real time applications. The Internet research community has devoted much effort to designing an integrated services Internet architecture, which is an architecture capable of supporting real time applications as well as data applications (see, for example, [4, 6, 7, 8, 10, 15, 18] and references therein for a small sampling of the research in this area) In a culmination of these efforts, the Internet Engineering Task Force (IETF) recently promoted to Proposed Standard extensions to the Internet architecture that will enable it to support reservations, in which resources ....

Sally Floyd. Comments on measurement-based admissions control for controlled-load services. submitted to CCR, July 1996.

....parameter based and three measurementbased for controlled load service. The parameter based admission control ensures that the sum of reserved resources is bounded by capacity. The three measurementbased algorithms are based on measured bandwidth, acceptance region [9] and equivalent bandwidth [7]. We use simulationon several network scenarios to evaluate the link utilization and adherence to service commitment achieved by these four algorithms. Keywords: Admission Control, Quality of Service (QOS) Internet. 1 Introduction The role of any admission control algorithm is to ensure that ....

....of this algorithm ensures that the measured instantaneous load plus the peak rate of a new flow is below the acceptance region. The measured load used in this scheme is not artificially adjusted upon admittance of a new flow. For flows described by a token bucket filter #r;b#but not peak rate, [7] derives their peak rates (bp) from the token bucket parameters using the equation: b p = r b=U; 3) where U is a user defined averaging period. We have adopted the same scheme to be used with the acceptance region algorithm. If a flow is rejected, the admission control algorithm does not ....

[Article contains additional citation context not shown here]

S. Floyd. "Comments on Measurement-based Admissions Control for Controlled-Load Service". Submitted to Computer Communication Review, 1996. URL ftp://ftp.ee.lbl.gov/papers/admit.ps.Z.

....multiplexer s bu er dynamics, even in the case of heterogeneous ows with highly bursty (temporally correlated) trac. Moreover, as the envelope does not employ a central limit Gaussian trac approximation, it can characterize scenarios with moderate statistical multiplexing gains: as detailed in [11], this is important in link sharing scenarios with a potentially moderate number of ows per trac class. Next, exploiting properties of this measured aggregate trac envelope, we devise a new envelope based MBAC algorithm as follows. First, subject to the (temporary) assumption that the past ....

....measurement time scale is a fundamental one to MBAC, and further discussions of its proper setting can be found in [4] 16] 18] for example. III. Simulation Experiments In this section, we evaluate the performance of the new algorithm for measurement based admission control and compare with [11] and [18] The workload consists of a set of twenty 30 minute traces of MPEG and JPEG compressed video from [26] In addition, we perform simulations using heavy tailed on o sources, which also form a long range dependent trac ow in aggregate. With this collection of traces and an implementation ....

[Article contains additional citation context not shown here]

S. Floyd. Comments on measurement-based admissions control for controlled-load services, July 1996. Lawrence Berkeley Laboratory Technical Report.

....accurately estimates a priori the level of service failures that will result. The other is to achieve the highest possible utilization for a given level of service failures. Several measurement based admission control algorithms have been proposed in the literature (see, for example, 7] 10] [11], 13] 14] 15] 16] 17] 19] 20] 21] and they implicitly or explicitly seek to achieve one or both of these design goals. The proposed algorithms, although embracing similar goals, differ in four important ways. First, some algorithms are principled, based on solid mathematical ....

....if the sum of the token rate of the new flow and the estimated rate of existing flows is less than a utilization target times the link bandwidth. A time window estimator is used to derive the estimated rate of existing flows. Hoeffding Bounds (HB) The admission control algorithm described in [11] computes the equivalent bandwidth for a set of flows using the Hoeffding bounds. A new flow is admitted if the sum of the peak rate of the new flow and the measured equivalent bandwidth is less than the link utilization. An exponential averaging measurement mechanism is used to produce the load ....

[Article contains additional citation context not shown here]

FLOYD, S. Comments on measurement-based admissions control for controlled-load services. Technical report, Lawrence Berkeley Laboratory, July 1996.

....(i) 1 log E e X (i) Then if there exists an s such that L X i=1 s (i) s; then (8.2) is ensured. Hui [24] obtained upper bounds on this type of e ective bandwidth by large deviation technics and Dueld [11] see Theorem 5.6) generalized his results. Floyd [14] uses the Hoe ding inequality to get bounds containing only the peak rates and the aggregate mean rate. Tur anyi, Veres and Ol ah [46] improved this result by the Cherno bound. Based on the latter two bounds, measurement based CAC algorithm can be provided, where the users declare their peak rate ....

S. Floyd. Comments on measurements based admission control for controlled-load services. Technical report, July 1996.

....measured sum approach, which bases admission decisions on an estimate of the mean aggregate arrival rate. Gibbens et al. 12] 11] study Chernoff bound based admission rules. They assume on off traffic and consider a tangent on the effective bandwidth function in their admission rule. Floyd [10] as well as Brichet and Simonian [4] study measurement based admission control based on the Hoeffding bound. They employ an exponential weighted moving average measurement mechanism. All these approaches have structural similarities, which are studied by Jamin and Shenker [17] Roughly speaking, ....

S. Floyd. Comments on measurement--based admission control for controlled--load services. submitted, July 1996. available at http://www-nrg.ee.lbl.gov:80/floyd.

....of Cambridge, Cambridge, CB2 1SB U.K. P. B. Key is with Microsoft Research, Cambridge, CB2 3NH,U.K. S. Zachary is with Heriot Watt University, Edinburgh, EH14 4AS, U.K. vided the cell scale and burst scales are suitably reinterpreted. MBAC ideas also found a home in the Internet community [6], 7] 8] and have been proposed as a way of limiting the number of flows or connections, where a flow can be a general transfer of data between a source and an endpoint or endpoints. It is worthwhile recapping the lessons that have been learned from the measurement based research. First, ....

S. Floyd, "Comments on measurement-based admissions control for controlled-load services." http://www.aciri.org/floyd/admit.html, July 1996.

....is derived from the logarithmic moment generating function, but use the actual logarithmic moment generating function in our admission rule. The LD admission rule thus takes the mean as well as the higher moments of the measured aggregate arrivals into account. Hoeffding Bound Approach Floyd [18] studies measurement based admission control based on the Hoeffding bound [19, 20] The Hoeffding bound is a Chernoff style bound for sums of bounded, independent random variables. Recall that U j ; j = 1; J , are steady state random variables denoting the rate at which connection j ....

....A3 of [15] for details. Measurement based admission control based on the Hoeffding 17 bound is also studied by Brichet and Simonian [21] They derive a tighter bound on the effective bandwidth of an on off stream 1 s U j (s) by considering a series expansion of U j (s) for small s. Floyd [18] as well as Brichet and Simonian [21] employ an exponential weighted moving average measurement mechanism. Let x denote the estimate of the aggregate arrivals in an averaging period of length T , i.e. x denotes the estimate of T P J j=1 r j . The estimate x is updated using the recursion x ....

S. Floyd. Comments on measurement--based admission control for controlled--load services. submitted, July 1996. available at http://www-nrg.ee.lbl.gov:80/floyd.

....packets into bursts of size b followed by a quiescent period of time b r . Alternatively, one could probe at some rate r 0 that is a function of the r and b. For instance, some measurement based admission control (MBAC) algorithms use an effective peak rate that is a function of r and b (see [9]) this value could be used as the probing rate. As we argued in Section 2, the admission control threshold should probably be a uniform standard. 13 For most of our simulations we assume that all flows use the same threshold ffl. However, we do run one test where the ffl s are different to ....

S. Floyd. Comments on measurement-based admissions control for controlled-load services, July 1996. Lawrence Berkeley Laboratory Technical Report.

....for audio, video, and other so called real time applications. The Internet research community has devoted much effort to designing an integrated services Internet architecture, which is an architecture capable of supporting real time applications as well as data applications (see, for example, [3, 5, 6, 7, 9, 14, 17] and references therein for a small sampling of the research in this area) In a culmination of these efforts, the Internet Engineering Task Force (IETF) recently promoted to Proposed Standard extensions to the Internet architecture that will enable it to support reservations, in which resources ....

Sally Floyd. Comments on measurement-based admissions control for controlled-load services. submitted to CCR, July 1996.

....measurements. Using ongoing measurements of load in making admission decisions is suggested, but not fully developed nor explored, in [OON88] Several recent papers, such as [SS91, AS94] use measurement to learn the parameters of certain assumed traffic distributions. The authors of [DJM96, Flo96a] use measurement of existing traffic in their calculation of equivalent bandwidth. In references [Hir91, CLG95] a neural network is used for dynamic bandwidth allocation. In [LCH95] the authors use pre computed low frequency of flows to allocate bandwidth dynamically by renegotiation. Hardware ....

....was later generalized in reference [Kel91] to handle resource with buffer. The theory of large deviation bounds the probability of rare events occuring. In this case, the rare event is S t . The approximationn in references [Hui88, Kel91] are based on the Chernoff s bound, while the one in [Flo96a] is based on the Hoeffding s bound. The Hoeffding s bound does not require that X i;t be independent of X i;t ffi . Equivalent bandwidth computed using the Hoeffding s bound is given by: CH ( fp i g 1in ; ffl) s ln(1=ffl) P n i=1 (p i ) 2 2 ; 2.13) where is the average arrival rate ....

[Article contains additional citation context not shown here]

S. Floyd. "Comments on Measurement-based Admissions Control for Controlled-Load Service". Submitted to Computer Communication Review, 1996. URL ftp://ftp.ee.lbl.gov/papers/admit.ps.Z.

.... with rate variations over multiple time scales, which are not adequately characterized by standard traffic models such as the leaky bucket [GW94, GKT, LPP94] 2 An alternative technique for supporting applications with ill specified traffic characteristics is a measurement based service [DJM97, Flo96, TG97, HW96, SMT96, JDSZ97, JSD97, CKT96, GKK95, GK97, CLM 97] by basing admission control decisions on measured values of traffic parameters rather than a priori client specified guesses, the effects of mistaken client traffic characterizations are largely alleviated, as is the need for a ....

....numbers of flows 4. temporally correlated aggregate traffic. Our MBAC algorithm described previously addresses this scenario. Furthermore, our experiments of Section 3 indicate that the algorithm accurately allocates resources in these environments. In contrast, previous approaches such as [DJM97, Flo96, GKK95, TG97, HW96, SMT96, JSD97] consider simpler scenarios which do not easily generalize to the scenario described above. For example, if traffic flows are homogeneous on off Bernoulli sources with identical peak rates, the on rate and on probability can be efficiently estimated as described ....

[Article contains additional citation context not shown here]

S. Floyd. Comments on measurement-based admissions control for controlled-load services, July 1996. Lawrence Berkeley Laboratory Technical Report.

....these characteristics may be unknown in advance and unpredictable in the future. This makes traffic modeling extremely difficult. Since integrated services networks [17] encourage a diversity of applications, it is believed that tightly modeling the traffic is neither necessary nor plausible [33]. Second, the admission of a new flow under parameter based approach usually implies a resource reservation to support the requested QoS of the new flow. If applied on multimedia traffic which is usually bursty this isolation of resources could result in low network utilization. A very ....

.... decent network utilization as well as decent predictive service commitment [22] It is well believed that most of the real time applications have sufficient adaptiveness to survive occasional service deterioration, especially if the service deterioration is controlled to happen infrequently [17, 21, 22, 33]. Many multimedia applications should thus benefit from measurement based admission control because of the high utilization it provides, its simplicity, and the diversity of its applicability. 5.2 Admission control for OCP A In [22] a measurement based admission control algorithm is introduced ....

S. Floyd. Comments on measurement-based admissions control for controlled-load services. Submitted to Computer Communication Review, from: ftp://ftp/ee/lbl.gov/papers/admit.ps.Z, July 1996.

....3. A III (n) does not require information on measurements of individual connection classes, but only on an aggregate flow measurement, that can be very useful. A slight modification of (32) leads to the Hoeffding bound. For a detailed analysis of measurement CAC based on the Hoeffding bound see [4]. 4. A IV (n) does not require information on the number of connections in each class, nether information on the total number of connections. All equations for admission regions contains a free parameter s. This parameter determines the trade off between link utilisation and loss ratio in the ....

.... compare the measured sum algorithm (this algorithm is introduced in [19] and is not based on the large deviation techniques) the Hoeffding bounds algorithm (this algorithm can be derived from the admission region A III (n) nevertheless Jamin and Shenker use the modified version described in [4]) and two algorithms that uses the above given admission regions, namely the Tangent at peak algorithm that uses the admission region A I (n) and the Tangent at origin algorithm that uses the admission region A IV (n) All the algorithms use the same ON OFF traffic model. There is an Internet ....

[Article contains additional citation context not shown here]

Floyd, S. . Comments on measurement-based admission control for controlled-load services. Lawrence Berkeley National Laboratory, 1996.

....) maximum measurement over whole T ; S ; if S ; r ff ; when adding a new flow ff; where S is the average rate over S averaging period. 3. 4 MBAC for Controlled Load Services Floyed introduces the measurement based admission control procedures for controlled load service in [2] (refer to [13] for details on controlled load service) The MBAC algorithm in [2] is given by employing the approach of equivalent capacity for a class of aggregated traffic. It uses measurement based estimates of the average aggregate arrival rate along with the peak rate of individual flows. ....

....a new flow ff; where S is the average rate over S averaging period. 3. 4 MBAC for Controlled Load Services Floyed introduces the measurement based admission control procedures for controlled load service in [2] refer to [13] for details on controlled load service) The MBAC algorithm in [2] is given by employing the approach of equivalent capacity for a class of aggregated traffic. It uses measurement based estimates of the average aggregate arrival rate along with the peak rate of individual flows. Let C(ffl) be the equivalent capacity such that the stationary arrival rate for the ....

S. Floyd, "Comments on measurement-based admission control for controlled-load servides, " Lawrence Berkeley National Laboratory Technical Report, Jul. 1996.

....r i . The time constant for this is, t = Gamma1= ln(1 Gamma w) A, where A is the length of each measurement interval. The time constant t should be as long as the interval from when a connection is admitted, until the measurements are likely to actually reAEect traOEc from that connection [20]. If it is C p e n i = C Request Call Yes Measurement On line Call Release Call Released Reject Call Admit Call No CAC Idle CAC Estimate Average Arrival Rate Figure 5: The measurement based CAC process in SDL diagram form. too short, the potential traOEc from newly admitted connections will ....

.... Gamma6 . This is very similar in principle to the eoeective bandwidth method described earlier. However, within a measurement based framework, certain approximations are necessary because of often small or moderate number of connections or types of connections. In [24] the Chernooe bounds and in [20] Hoeoeding bounds are used to determine the probability that the bandwidth required by connected AEows is greater than the link capacity must stay below a given value using the measured average aggregate arrival rate of existing connections and the peak rate of the requested connection and is used ....

Floyd, S., Comments on Measurement-based Admissions Control for Controlled-Load Services, Computer Communication Review 1996. (ftp://ftp.ee.lbl.gov/papers/admit.ps.Z )

....a knowledge of the bandwidth requirement. Finally, it is impossible to tell what types of traffic it may be necessary to transmit in the future, and it should not be required that each new traffic type be submitted to a complex modeling process in advance of transmission. An alternative approach [8, 14, 15, 21, 22, 23] is to attempt to measure the bandwidth requirement more directly. This avoids the problem of requiring new traffic types to specify a parameterised model in advance and removes the estimation of redundant information. Perhaps most importantly, this approach requires very little declared ....

Sally Floyd. Comments on measurement based admission control for controlled-load services. Submitted to CCR, July 1996.

....in the network, and the cost of characterising the sources. Our approach imposes minimal requirements on the sources whilst permitting a high degree of statistical multiplexing gain in the network. These arguments have been recognised by several other authors, for example Jamin [22, 21] and Floyd [23]. Jamin et al. proposing a measurement based admission control algorithm for the IETF Predictive service, proposes substituting conservative delay bounds, computed from the worst case traffic which could be transmitted through a user parameterised token bucket, with estimated bounds computed from ....

....algorithm for the IETF Predictive service, proposes substituting conservative delay bounds, computed from the worst case traffic which could be transmitted through a user parameterised token bucket, with estimated bounds computed from simple measurements of the source activity. Floyd s approach [23] uses Hoeffding bounds to compute the equivalent capacity of the sources, viewed as a sum of random variables; based on this, she proposes a simple admission control procedure for the IETF Controlled Load service. The algorithm relies on policed source peak rates, together with an exponentially ....

Sally Floyd. Comments on measurement based admission control for controlled-load services. Submitted to CCR, July 1996.

....to a large and heterogeneous application base, and is very different from our approach to admission control that is based on ongoing measurements. In references [SS91, AS94] the authors use measurement to learn the parameters of certain assumed traffic distributions. The authors of [DJM96, Flo96] use measurement of existing traffic in their calculation of equivalent bandwidth, providing load, but not delay, bound. In references [Hir91, CLG95] a neural network is used for dynamic bandwidth allocation. In [LCH95] the authors use pre computed low frequency of flows to renegotiate ....

S. Floyd. "Comments on Measurement-based Admissions Control for Controlled-Load Service". Submitted to Computer Communication Review, 1996. URL ftp://ftp.ee.lbl.gov/papers/admit.ps.Z.

....to the analysis of a family of simple and robust measurement based admission controls. A subsidiary aim of the paper is to shed light on the relationship between the admission control proposed for ATM networks by Gibbens et al. [9] and that proposed for controlled load Internet services by Floyd [7]. We shall see that their common origin in Chernoff bounds allows the definition of a simple and general family of admission controls, capable of tailoring for several implementation scenarios. 1. Introduction There is by now a fairly good understanding of the behaviour of a queue whose arrival ....

....Research Fellowship and both authors the EPSRC (Grant GR J31896) for computing equipment. between time scales, and there remains much work to be done. Jamin and Shenker [11] have conducted a simulation based comparison of several measurement based admission control algorithms, including [7] and [9] We view the approach of this paper to be complementary to the simulation approach: each has its strengths and weaknesses, and a fuller understanding of measurement based admission control will require contributions from theory, modelling, simulation and experiment. For several other ....

[Article contains additional citation context not shown here]

S. Floyd. Comments on measurement-based admission control for controlled-load services. Lawrence Berkeley National Laboratory, July 1996.

....perform experiments with lower link rates and find that even in such cases where limited statistical multiplexing gains are available such as in a link sharing environment [7] the MBAC algorithm correctly restricts the admissible region to lower utilizations. Previous approaches to MBAC such as [4, 6, 9, 10, 11, 17] have employed a number of measurement methodologies to characterize traffic including instantaneous peak rate, the instantaneous rate s mean and variance or moment generating function, and per flow statistics. In addition, a number of theoretical techniques have been applied to study various ....

....measuring and controlling the average bandwidth utilization and the experienced maximal queueing delay D to target values. In our experiments, we use the recommended performance tuning parameters of [11] with v set to 0.9, 2, S = 1=24 sec, and T = 3 sec. We also compare with Floyd s approach [6], which, like the algorithm proposed in this paper, addresses scenarios with moderate numbers of multiplexed traffic flows such as in a link sharing environment in which the available capacity is partitioned to support a number of services and traffic classes. The approach is based on the ....

[Article contains additional citation context not shown here]

S. Floyd. Comments on measurement-based admissions control for controlled-load services, July 1996. Lawrence Berkeley Laboratory Technical Report.

....network clients to accurately characterize their traffic in advance. It therefore appears that if a refined traffic model is not used, statistical services would yield such low resource utilization for bursty traffic flows that either renegotiated services [22, 52] or measurement based services [18] must be used instead. While such services have their own merits as discussed in the respective works, they unfortunately cannot provide statistical QoS guarantees per se. Moreover, a renegotiated service requires increased signaling overhead and a measurement based service must successfully make ....

S. Floyd, "Comments on Measurement-based Admissions Control for Controlled-Load Services," July 1996. Lawrence Berkeley Laboratory Technical Report.

....only by a peak rate and a token bucket filter. Hence in this paper we only consider MBAC s that do not require precise source characterization. Aside from our own ( JDSZ97] several MBAC s requiring only a token bucket filter description of sources have been proposed in the literature [GKK95, Flo96, GK97] A number of studies have also been conducted to explore alternative measurement processes and their implementations in hardware [DJM97, C 91, WCKG94] The more recent works on MBAC s ( JSD97, GK97] consisted of comparative studies of several different MBAC schemes. In [JSD97] we ....

....point, we study the full range of achievable performance of each MBAC. 1 We call the range of achievable utilizations and observed loss rates the loss load curve. The study reported in [GK97] is similar in spirit to ours. Through formal analysis, the authors showed that the MBAC equations in [Flo96] and [GKK95] belong to the same class: tangents on the exact equivalent bandwidth curve computed using large deviation analysis. The authors further conducted numerical investigations on the trade off between loss rate and network utilization when different MBAC s, reflecting tangents at ....

[Article contains additional citation context not shown here]

S. Floyd. "Comments on Measurement-based Admissions Control for Controlled-Load Service". Submitted to Computer Communication Review, 1996. URL ftp://ftp.ee.lbl.gov/papers/admit.ps.Z.

....service) is whether or not to offer more than one priority level of service; that is, whether to have multiple levels of scheduling priority within the Controlled Load service. 3 The current service definition offers only a 1 The literature is far too vast to review here, but see [4, 5, 6, 7, 9, 15, 18] and references therein for a few representative examples. 2 Controlled Load service is a relaxed real time service that provides low delay and low loss, but does not provide delay bounds. 3 The analogous question remains even if real time applications are supported by a best effort network ....

Sally Floyd. Comments on measurement-based admissions control for controlled-load services. submitted to CCR, July 1996.

No context found.

S. Floyd, Comments on measurement-based admissions control for controlled-load services, LBNL, Tech. Rep., 1996.

No context found.

S. Floyd. Comments on measurement-based admissions control for controlled-load services. Technical report, LBNL, 1996.

No context found.

Floyd, S., "Comments on Measurement-based Admissions Control for Controlled Load Services". Technical report, Lawrence Berkeley Laboratory, July 1996.

No context found.

S. Floyd. Comments on measurement-based admission control for controlled load services. Technical Report, Lawrence Berkeley Laboratory, July 1996.

No context found.

S. Floyd. Comments on measurement-based admission control for controlled load services. Technical Report, Lawrence Berkeley Laboratory, July 1996.

No context found.

S. Floyd, Comments on Measurement-based Admission Control for Controlled Load Services, draft submitted to ACM Computer Communication Review, (also available at: http://wwwnrg. ee.lbl.gov/floyd/), July 1996.

No context found.

S. Floyd. Comments on measurement-based admission control for controlled-load services. Technical report, Lawrence Berkeley National Laboratory, July 1996.

No context found.

S. Floyd. Comments on measurement-based admission control for controlled-load services. Technical report, Lawrence Berkeley National Laboratory, July 1996.

No context found.

S. Floyd, Comments on measurement-based admission control for controlled load services, Technical Report, Lawrence Berkeley Laboratory, July 1996.

No context found.

S. Floyd. Comments on measurement-based admissions control for controlled-load services, July 1996. Lawrence Berkeley Laboratory Technical Report.

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