## A measurement-based admission control algorithm for integrated services packet networks (1997)

### Cached

### Download Links

- [sigcomm.org]
- [conferences.sigcomm.org]
- [www.cs.columbia.edu]
- [www.cs.ucla.edu]
- [catarina.usc.edu]
- [speed.cis.nctu.edu.tw]
- [irl.eecs.umich.edu]
- [www.cs.umn.edu]
- [irl.eecs.umich.edu]
- [www.eecs.umich.edu]
- [irl.eecs.umich.edu]
- [ftp.usc.edu]
- [www.eecs.umich.edu]
- [netweb.usc.edu]
- [qosip.tmit.bme.hu]
- DBLP

### Other Repositories/Bibliography

Venue: | IEEE/ACM TRANSACTIONS ON NETWORKING |

Citations: | 309 - 10 self |

### BibTeX

@INPROCEEDINGS{Jamin97ameasurement-based,

author = {Sugih Jamin and Peter B. Danzig and Scott J. Shenker and Lixia Zhang},

title = {A measurement-based admission control algorithm for integrated services packet networks },

booktitle = {IEEE/ACM TRANSACTIONS ON NETWORKING},

year = {1997},

pages = {56--70},

publisher = {}

}

### Years of Citing Articles

### OpenURL

### Abstract

Many designs for integrated service networks offer a bounded delay packet delivery service to support real-time applications. To provide bounded delay service, networks must use admission control to regulate their load. Previous work on admission control mainly focused on algorithms that compute the worst case theoretical queueing delay to guarantee an absolute delay bound for all packets. In this paper we describe a measurement-based admission control algorithm for predictive service, which allows occasional delay violations. We have tested our algorithm through simulations on a wide variety of network topologies and driven with various source models, including some that exhibit long-range dependence, both in themselves and in their aggregation. Our simulation results suggest that, at least for the scenarios studied here, the measurement-based approach combined with the relaxed service commitment of predictive service enables us to achieve a high

### Citations

2153 |
Time Series Analysis for Forecasting and Control
- Box, Jenkins
- 1970
(Show Context)
Citation Context ...set of weights, the degree of integration, and the moving average process order with the corresponding set of weights, it also requires an innovation with a Gaussian marginal distribution (see, e.g., =-=[4]-=-, [21]). We first generate a normally distributed innovation with mean and standard deviation packets. If the minimum of the fARIMA output is less than zero, we shift the whole series byJAMIN et al.:... |

1962 | On the self-similar nature of Ethernet traffic
- Leland, Taqqu, et al.
- 1994
(Show Context)
Citation Context ...call holding time (CHT, which we will also call “flow duration” or “flow lifetime”). One of these is the two-state Markov process used widely in the literature. Recent studies ([6], [13], [18], [28], =-=[30]-=-, [35]) have shown that network traffic often exhibits longrange dependence (LRD), with the implications that congested periods can be quite long and a slight increase in number of active connections ... |

1807 | Generalized Processor Sharing Approach to Flow Control in Integrated Services Networks: the Multiple Node Case
- Parekh, Gallager
- 1994
(Show Context)
Citation Context ...ctive traffic, we first need a model for the worst-case delay of priority queues. Cruz, in [9], derived a tight bound for the worst-case delay, of priority queue level Our derivation follows Parekh’s =-=[34]-=-, which is a simpler, but looser, bound for that assumes small packet sizes, i.e., the transmission time of each packet is sufficiently small (as compared to other delays) and hence can be ignored. Th... |

1574 | Wide-area Traffic: The Failure of Poisson Modeling
- Paxson, Floyd
- 1995
(Show Context)
Citation Context ...olding time (CHT, which we will also call “flow duration” or “flow lifetime”). One of these is the two-state Markov process used widely in the literature. Recent studies ([6], [13], [18], [28], [30], =-=[35]-=-) have shown that network traffic often exhibits longrange dependence (LRD), with the implications that congested periods can be quite long and a slight increase in number of active connections can re... |

801 |
Statistics for long-memory processes
- Beran
- 1994
(Show Context)
Citation Context ...g nonnegative series is also used in [1]. Note that this shifting process constrains the maximum value of the generated series to be always twice its average. The Whittle maximum likelihood estimator =-=[3]-=- confirms that our shifting, cropping, and overlaying of the fARIMA generated series does not destroy its long-range dependence. Fig. 3 shows a packet-arrival depiction of an ON/OFF source in the cont... |

664 | A scheme for real-time channel establishment in wade-area networks
- Ferrari, Verma
- 1990
(Show Context)
Citation Context ...When a flow requests real-time service, it must characterize its traffic so that the network can make its admission control decision. Typically, sources are described by either peak and average rates =-=[15]-=- or a filter like a token bucket [33]; these descriptions provide upper bounds on the traffic that can be generated by the source. The traditional real-time service provides a hard or Manuscript recei... |

659 | Self-similarity through high-variability: statistical analysis of Ethernet LAN traffic at the source level
- Willinger, Taqqu, et al.
- 1995
(Show Context)
Citation Context ...arameter less than two results in data with infinite variance. The Pareto location parameter is Each Pareto-ON/OFF source by itself does not generate LRD series. However, the aggregation of them does =-=[39]-=-. 3) LRD—Fractional ARIMA: We use each number generated by the fractional autoregressive integrated moving average (fARIMA) process [22] as the number of fixed-size packets to be sent back to back in ... |

606 |
A Calculus for Network Delay, Part I : Network Elements in Isolation
- Cruz
- 1991
(Show Context)
Citation Context ...approximations. A. Worst-Case Delay: Predictive Service To compute the effect of a new flow on existing predictive traffic, we first need a model for the worst-case delay of priority queues. Cruz, in =-=[9]-=-, derived a tight bound for the worst-case delay, of priority queue level Our derivation follows Parekh’s [34], which is a simpler, but looser, bound for that assumes small packet sizes, i.e., the tra... |

539 | Supporting real-time applications in an integrated services packet networks: architecture and mechanism
- Clark, Shenker, et al.
- 1992
(Show Context)
Citation Context ...A 90095 USA. S. J. Shenker is with the Xerox Palo Alto Research Center, Palo Alto, CA 94304-1314 USA. Publisher Item Identifier S 1063-6692(97)01637-3. absolute bound on the delay of every packet; in =-=[10]-=- and [15], this service model is called guaranteed service. Admission control algorithms (ACA’s) for guaranteed service use the a priori characterizations of sources to calculate the worst-case behavi... |

497 | Analysis, modeling and generation of self-similar VBR video traffic
- GARRETT, WILLINGER
(Show Context)
Citation Context ...ON time and call holding time (CHT, which we will also call “flow duration” or “flow lifetime”). One of these is the two-state Markov process used widely in the literature. Recent studies ([6], [13], =-=[18]-=-, [28], [30], [35]) have shown that network traffic often exhibits longrange dependence (LRD), with the implications that congested periods can be quite long and a slight increase in number of active ... |

450 |
Equivalent capacity and its application to bandwidth allocation in high-speed networks
- Guerin, Ahmadi, et al.
(Show Context)
Citation Context ...andwidth that is larger than its average rate but less than its peak rate. In most cases, the equivalent bandwidth is computed based on a statistical model [23], [37] or on a fluid flow approximation =-=[16]-=-, [27] of traffic. 1 If one can precisely characterize traffic a priori, this approach will increase network utilization. However, we think it will be quite difficult, if not impossible, to provide ac... |

394 | Specification of the controlled-load network element service. Internet Request For Comments RFC 2211, Internet Engineering Task Force
- Wroclawski
- 1997
(Show Context)
Citation Context ...x.com CSE-TR-333-97 April 1997 Abstract The Internet Engineering Task Force (IETF) is considering the adoption of the controlled-load service, a real-time service with very relaxed service guarantees =-=[Wro95]-=-. Measurement-based admission control algorithms (MBAC’s), as opposed to the more conservative worst-case parameter-based methods, were expressly designed to achieve high levels of network utilization... |

286 |
Long-range dependence in variable-bit-rate video traffic
- Beran
- 1995
(Show Context)
Citation Context ...n of their ON time and call holding time (CHT, which we will also call “flow duration” or “flow lifetime”). One of these is the two-state Markov process used widely in the literature. Recent studies (=-=[6]-=-, [13], [18], [28], [30], [35]) have shown that network traffic often exhibits longrange dependence (LRD), with the implications that congested periods can be quite long and a slight increase in numbe... |

191 | Effective bandwidths at multi-class queues
- Kelly
- 1991
(Show Context)
Citation Context ...th that is larger than its average rate but less than its peak rate. In most cases, the equivalent bandwidth is computed based on a statistical model [23], [37] or on a fluid flow approximation [16], =-=[27]-=- of traffic. 1 If one can precisely characterize traffic a priori, this approach will increase network utilization. However, we think it will be quite difficult, if not impossible, to provide accurate... |

181 |
Resource Allocation for Broadband Networks
- Hui
- 1988
(Show Context)
Citation Context ...proach, each flow is allotted an effective bandwidth that is larger than its average rate but less than its peak rate. In most cases, the equivalent bandwidth is computed based on a statistical model =-=[23]-=-, [37] or on a fluid flow approximation [16], [27] of traffic. 1 If one can precisely characterize traffic a priori, this approach will increase network utilization. However, we think it will be quite... |

144 | Notes on effective bandwidths - Kelly - 1996 |

125 | Space-time modelling with longmemory dependence: Assessing Ireland’s wind power resource (with discussion
- Haslett, Raftery
- 1989
(Show Context)
Citation Context ... not generate LRD series. However, the aggregation of them does [39]. 3) LRD—Fractional ARIMA: We use each number generated by the fractional autoregressive integrated moving average (fARIMA) process =-=[22]-=- as the number of fixed-size packets to be sent back to back in each ON period. Interarrivals of ON periods are of fixed length. For practical programming reasons, we generate a series of 15 000 fARIM... |

117 |
Modeling Persistence in Hydrological Time Series Using Fractional Differencing
- Hosking
- 1984
(Show Context)
Citation Context ...f weights, the degree of integration, and the moving average process order with the corresponding set of weights, it also requires an innovation with a Gaussian marginal distribution (see, e.g., [4], =-=[21]-=-). We first generate a normally distributed innovation with mean and standard deviation packets. If the minimum of the fARIMA output is less than zero, we shift the whole series byJAMIN et al.: A MEA... |

80 |
A Decision-Theoretic Approach to Call Admission Control in ATM Networks
- Gibbens, Kelly, et al.
- 1995
(Show Context)
Citation Context ...this paper has been extended in [12] to support advance reservations. Degermark et al. [12] have also replicated some of our results on their independently developed network simulator. The authors of =-=[17]-=- and [20] use measurements to determine admission control, but the admission decisions are precomputed based on the assumption that all sources are exactly described by one of a finite set of source m... |

79 | Statistical analysis of CCSN/SS7 traffic data from working CCS subnetworks
- Duffy, Mcintosh, et al.
- 1994
(Show Context)
Citation Context ...their ON time and call holding time (CHT, which we will also call “flow duration” or “flow lifetime”). One of these is the two-state Markov process used widely in the literature. Recent studies ([6], =-=[13]-=-, [18], [28], [30], [35]) have shown that network traffic often exhibits longrange dependence (LRD), with the implications that congested periods can be quite long and a slight increase in number of a... |

74 | Comments on measurement-based admissions control for controlled-load services
- Floyd
- 1996
(Show Context)
Citation Context ...eed some level over the duration of their flow, no matter how short the flow; however, we contend that the provider cannot realistically assure that excessive violations will never occur. of [11] and =-=[14]-=- use measurement of existing traffic in their calculation of equivalent bandwidth, providing load, but not delay, bound. In [8] and [19], a neural network is used for dynamic bandwidth allocation. In ... |

68 | Pacifici G., A separation principle between scheduling and admission control for broadband switching
- Hyman, Lazar
- 1993
(Show Context)
Citation Context ...r has been extended in [12] to support advance reservations. Degermark et al. [12] have also replicated some of our results on their independently developed network simulator. The authors of [17] and =-=[20]-=- use measurements to determine admission control, but the admission decisions are precomputed based on the assumption that all sources are exactly described by one of a finite set of source models. Th... |

66 | Predictive Dynamic Bandwidth Allocation for Efficient Transport of RealTime VBR Video over ATM
- Chong, Li, et al.
- 1995
(Show Context)
Citation Context ...stically assure that excessive violations will never occur. of [11] and [14] use measurement of existing traffic in their calculation of equivalent bandwidth, providing load, but not delay, bound. In =-=[8]-=- and [19], a neural network is used for dynamic bandwidth allocation. In [29], the authors use precomputed low frequency of flows to renegotiate bandwidth allocation. Hardware implementation of measur... |

61 | Providing end-to-end statistical performance guarantees with bounding interval dependent stochastic models
- Zhang, Knightly
- 1994
(Show Context)
Citation Context ...ed service inevitably results in low utilization [40]. Higher network utilization can be achieved by weakening the reliability of the delay bound. For instance, the probabilistic service described in =-=[41]-=- does not provide for the worst-case scenario, instead it guarantees a bound on the rate of lost/late packets based on statistical characterization of traffic. In this approach, each flow is allotted ... |

59 |
Admission Control Algorithms for Predictive Real-Time Service
- Jamin, Shenker, et al.
- 1992
(Show Context)
Citation Context ... incur delay violations gives admission control a great deal more flexibility, and is the chief advantage of predictive service. The measurement-based admission control approach advocated in [10] and =-=[26]-=- uses the a priori source characterizations only for incoming flows (and those very recently admitted); it uses measurements to characterize those flows that have been in place for a reasonable durati... |

57 | fmproving utilization for deterministic service in multimedia communication
- Zfrang, Ferrari
- 1994
(Show Context)
Citation Context ...ddition to the incoming one. Network utilization under this model is usually acceptable when flows are smooth; when flows are bursty, however, guaranteed service inevitably results in low utilization =-=[40]-=-. Higher network utilization can be achieved by weakening the reliability of the delay bound. For instance, the probabilistic service described in [41] does not provide for the worst-case scenario, in... |

56 | A Scheduling Service Model and a Scheduling Architecture for an Integrated Services Packet Network.” Typescript, Xeorox PARC technical report
- Shenker, Clark, et al.
- 1993
(Show Context)
Citation Context ... applications adapt to actual packet delays and are thus rather tolerant of occasional delay bound violations; they do not need an absolutely reliable bound. For these tolerant applications, [10] and =-=[36]-=- proposed predictive service, which offers a fairly, but not absolutely, reliable bound on packet delivery times. The ability to occasionally incur delay violations gives admission control a great dea... |

51 | On Resource Management and QoS Guarantees for Long Range Dependent Traffic
- Adas, Mukherjee
- 1995
(Show Context)
Citation Context ...stributed [35]. Fig. 3. ON/OFF traffic model with token-bucket filter. adding the absolute value of its minimum to every number in the series. This way of obtaining nonnegative series is also used in =-=[1]-=-. Note that this shifting process constrains the maximum value of the generated series to be always twice its average. The Whittle maximum likelihood estimator [3] confirms that our shifting, cropping... |

44 |
Modeling call holding time distributions for ccs network design and performance analysis
- Bolotin
- 1994
(Show Context)
Citation Context ...5 360 20 10 1.2 1.9 s 64 240 1 60 0 4.5e-5 1 125 8 13 1024 100 1.1e-2 34 100 160 0 220 16 256 16 160 [31]. The duration of for LRD sources, however, are taken from a lognormal distribution, following =-=[5]-=- and [13]. The interarrival times of all flows are exponentially distributed [35]. Fig. 3. ON/OFF traffic model with token-bucket filter. adding the absolute value of its minimum to every number in th... |

44 | Capacity Allocation and Network Control by Filtered Input Rate in High-Speed Networks
- Li, Chong, et al.
- 1995
(Show Context)
Citation Context ... use measurement of existing traffic in their calculation of equivalent bandwidth, providing load, but not delay, bound. In [8] and [19], a neural network is used for dynamic bandwidth allocation. In =-=[29]-=-, the authors use precomputed low frequency of flows to renegotiate bandwidth allocation. Hardware implementation of measurement mechanisms are studied in [7] and [38]. II. MEASUREMENT-BASED ADMISSION... |

37 |
Dynamic call admission control in ATM networks
- Saito, Shiomoto
- 1991
(Show Context)
Citation Context ..., each flow is allotted an effective bandwidth that is larger than its average rate but less than its peak rate. In most cases, the equivalent bandwidth is computed based on a statistical model [23], =-=[37]-=- or on a fluid flow approximation [16], [27] of traffic. 1 If one can precisely characterize traffic a priori, this approach will increase network utilization. However, we think it will be quite diffi... |

36 | The impact of the ATM concept on video coding - Verbiest, Pinoo, et al. |

33 | Advance reservations for predictive service
- Degermark, Kohler, et al.
- 1995
(Show Context)
Citation Context ..., does offering predictive service increase network utilization? Earlier versions of this work have been published in [25] and [26]. Incidentally, the work reported in this paper has been extended in =-=[12]-=- to support advance reservations. Degermark et al. [12] have also replicated some of our results on their independently developed network simulator. The authors of [17] and [20] use measurements to de... |

31 | A framework for bandwidth management in ATM networks – aggregate equivalent bandwidth estimation approach - Dziong, Juda, et al. - 1997 |

21 |
On defining, computing and guaranteeing quality-of-service in high-speed networks
- Nagarajan, Kurose
- 1992
(Show Context)
Citation Context ...ve service itself does not specify an acceptable level of delay violations. This is for two reasons. First, it is not particularly meaningful to specify a failure rate to a flow with a short duration =-=[32]-=-. Second, reliably ensuring that the failure rate never exceeds a particular level leads to the same worst-case calculations that predictive service was designed to avoid. Instead, the CSZ approach [1... |

20 |
NOGUCHI: “Flow control schemes and delay/loss tradeoff in ATM networks
- OHNISHI, OKADA, et al.
- 1988
(Show Context)
Citation Context ...e, it must characterize its traffic so that the network can make its admission control decision. Typically, sources are described by either peak and average rates [15] or a filter like a token bucket =-=[33]-=-; these descriptions provide upper bounds on the traffic that can be generated by the source. The traditional real-time service provides a hard or Manuscript received October 1995; revised September 1... |

19 | An Adaptive Algorithm for Measurement-based Admission Control
- Casetti, Kurose, et al.
- 1996
(Show Context)
Citation Context ...s 5e3 3 0.35, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.99, 0.999, 0.9999 5e3 4 0.7, 0.8, 0.9, 0.9417, 0.99, 0.999, 0.9999 6.5e3 4 0.8, 0.9, 0.9417, 0.999, 0.9999 Extending our work in [JDSZ95], the authors of =-=[CKT96]-=- suggest a feedback control mechanism to automatically tune the measurement parameters based on observed loss rate. The appropriate time scale over which these and parameters in the ssand sfu... |

17 | C.; On Long-Range Dependence in NSFNET Traffic - Klivanski, Mukherjee, et al. - 1994 |

15 |
Integration of ATM call admission control and link capacity control by distributed neural networks,” lEEE
- Hiramatsu
- 1991
(Show Context)
Citation Context ... assure that excessive violations will never occur. of [11] and [14] use measurement of existing traffic in their calculation of equivalent bandwidth, providing load, but not delay, bound. In [8] and =-=[19]-=-, a neural network is used for dynamic bandwidth allocation. In [29], the authors use precomputed low frequency of flows to renegotiate bandwidth allocation. Hardware implementation of measurement mec... |

9 |
A traffic control method for service quality assurance in an ATM network
- Abe, Soumiya
- 1994
(Show Context)
Citation Context ...e models. This approach is clearly not applicable to a large and heterogeneous application base, and is very different from our approach to admission control that is based on ongoing measurements. In =-=[2]-=- and [37], the authors use measurement to learn the parameters of certain assumed traffic distributions. The authors 2 A network provider might promise to give its customers their money back if the vi... |

8 | Specification of the Committed Rate Quality of Service. Internet-Draft - Baker, Guérin, et al. - 1996 |

7 |
Application of the Theory of Probability to Telephone Trunking Problems. Bell Syst
- Molina
- 1927
(Show Context)
Citation Context ... 50 2.1e-3 17 160 160 684 9 1 512 80 9.4e-5 1 160 160 POO1 POO2 fARIMA (f0:75g; 0.15, -) 64 256 2925 360 20 10 1.2 1.9 s 64 240 1 60 0 4.5e-5 1 125 8 13 1024 100 1.1e-2 34 100 160 0 220 16 256 16 160 =-=[31]-=-. The duration of for LRD sources, however, are taken from a lognormal distribution, following [5] and [13]. The interarrival times of all flows are exponentially distributed [35]. Fig. 3. ON/OFF traf... |

6 | private communication - Willinger - 2001 |

4 | Adaptive Source Routing of Real-Time Traffic - Breslau - 1995 |

4 | VariableBit -Rate Video Traffic and Long-Range Dependence - Beran, Sherman, et al. - 1995 |

3 | Equivalent Capacity and Its Application to Bandwidth Allocation in HighSpeed Networks - Naghshineh - 1991 |

3 |
Real-Time Traffic Esitmation
- Warfield, Chan, et al.
- 1994
(Show Context)
Citation Context ...dynamic bandwidth allocation. In [29], the authors use precomputed low frequency of flows to renegotiate bandwidth allocation. Hardware implementation of measurement mechanisms are studied in [7] and =-=[38]-=-. II. MEASUREMENT-BASED ADMISSION CONTROL FOR ISPN Our ACA consists of two logically distinct aspects. The first aspect is the set of criteria controlling whether to admit a new flow; these are based ... |

2 | is currently a Principal Scientist at the Xerox Palo Alto Research Center. He spent the 1983–1994 academic year at Cornell University - He |

2 | An Adaptive Algorithm for Measurement-basedAdmission Control in Integrated Services Packet Networks - Casetti, Kurose, et al. - 1996 |

2 |
et al. “Interconnection of Dual Bus MANs: Architecture and Algorithms for Bandwidth Allocation
- Conti
- 1991
(Show Context)
Citation Context ...sed for dynamic bandwidth allocation. In [29], the authors use precomputed low frequency of flows to renegotiate bandwidth allocation. Hardware implementation of measurement mechanisms are studied in =-=[7]-=- and [38]. II. MEASUREMENT-BASED ADMISSION CONTROL FOR ISPN Our ACA consists of two logically distinct aspects. The first aspect is the set of criteria controlling whether to admit a new flow; these a... |