D. E. Duffy, A. A. McIntosh, M. Rosenstein and W. Willinger, "Statistical analysis of CCSN/SS7 traffic data from working CCS subnetworks," IEEE J. Select. Areas Commun., vol. 12, no. 3, pp. 544 - 551, Apr. 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....15, 16] and to the introduction of the term self similar (or fractal) traffic. Since then, this feature has been discovered in many other traffic traces, such as Transmission Control Protocol (TCP) 10, 11] Motion Pictures Experts Group (MPEG) video [5] World Wide Web [3] and Signaling System 7 [4] traffic. An important characteristic of self similar traffic is its long range dependence, i.e. the existence of correlations over a broad range of time scales [2, 12] Most network traffic measurements have been performed on wired networks. The question arises whether the traffic in wireless ....

D. E. Duffy, A. A. McIntosh, M. Rosenstein and W. Willinger, "Statistical analysis of CCSN/SS7 traffic data from working CCS subnetworks," IEEE J. Select. Areas Commun., vol. 12, no. 3, pp. 544 - 551, Apr. 1994.

....at much lower link utilization compared to our earlier results. POO1 this is an on off source with the same averages as the EXP1 source. However, they are taken from a Pareto distribution. Flow lifetimes are taken from a lognormal distribution with a median of 300 seconds following [2] [9]. The aggregation of these sources produces traffic that is long range dependent [8] 25] Because of increased variability with these traffic sources, simulations were run for 20,000 seconds, with data for the first 10,000 seconds discarded. The resulting performance frontiers are shown in ....

DUFFY, D., MCINTOSH, A., ROSENSTEIN, M., AND WILLINGER, W. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks". IEEE Journal on Selected Areas in Communications 12, 3 (Apr. 1994), 544--551.

....burstiness, network traffic dynamics is also effected by the arrival pattern and duration of flows. We use exponentially distributed lifetimes with the Markov ON OFF source model, following[14] The duration of Pareto ON OFF sources, however, are taken from a lognormal distribution, following [3, 5]. The interarrival times of all flows are exponentially distributed [16] We use six instantiations of the above source models as summarized in Table 1. In the table, p = 1 means that after each OFF time, packets for the next ON period are transmitted back to back. The shape parameter of the ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, and W. Willinger. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks". IEEE Journal of Selected Areas in Communication, 12(3):544--551, Apr. 1994.

....model in our simulations. All of them are on off processes. They differ in the distribution 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 ( LTWW94, DMRW94, PF94, KM94, GW94, BSTW95] have shown that network traffic often exhibits long range dependence (lrd) with the implications that congested periods can be quite long and a slight increase in number of active connections can result in large increase in packet loss rate [PF94] Reference [PF94] ....

....and the mix of transport protocol, source model, token bucket filter, and service request ascribed to each flow. We ascribe exponentially distributed lifetimes to the exp model, following [Mol27] The duration of for lrd sources, however, are taken from a lognormal distribution, following [Bol94, DMRW94] The interarrival times of all flows are exponentially distributed [PF94] 4.3 Parameter Choices We chose six instantiations of the above three source models, as summarized in Table 2. In the table, p = 1 means that after each off time, packets for the next on period are transmitted back to ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, and W. Willinger. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks ". IEEE Journal of Selected Areas in Communication, 12(3):544-- 551, Apr. 1994.

....average rate is held constant, leading to a burstier source model. POO1 this is an on off source with the same averages as the EXP1 source. However, they are taken from a Pareto distribution. Flow lifetimes are taken from a lognormal distribution with a median of 300 seconds following [2] [9]. The aggregation of these sources produces traffic that is long range dependent [8] 25] STARWARS this source model is taken from a trace file produced by an MPEG encoding of the Star Wars motion picture [12] Each source starts from a random place within the trace file in order to avoid ....

DUFFY, D., MCINTOSH, A., ROSENSTEIN, M., AND WILLINGER, W. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks ". IEEE Journal on Selected Areas in Communications 12, 3 (Apr. 1994), 544--551.

....21] and the introduction of the term self similar (or fractal) traffic. Since then, this feature has been discovered in many other traffic traces, such as Transmission Control Protocol (TCP) 14 16] Motion Pictures Experts Group (MPEG) video [6] World Wide Web [4, 12] and Signaling System 7 [5] traffic. An important characteristic of self similar traffic is its longrange dependence, i.e. the existence of correlations over a broad range of time scales [2, 3, 17] Until now, most network traffic measurements were performed on wired networks. The question arises whether the traffic in ....

D. E. Duffy, A. A. McIntosh, M. Rosenstein and W. Willinger, "Statistical analysis of CCSN/SS7 traffic data from working CCS subnetworks," IEEE J. Select. Areas Commun., vol. 12, no. 3, pp. 544 - 551, Apr. 1994.

.... sources with LRD include video coders [10, 2] and ISDN terminals [15] Examples of network traffic (i.e. resulting from of the superposition of multiple sources) with LRD include Ethernet local area networks [12] and wide area packet networks such as the Internet [17, 11] or the CCS network [7]. It turns out that traffic with LRD from an indivdual source can be modeled using the familiar on off model widely used in conventional analysis. In this model, the source is active and transmits data at a constant rate (often referred to as the peak rate) during the on period, and it is ....

D. Duffy, A. McIntosh, M. Rosenstein, D. Wilson, "Statistical analysis of CCSN/SS7 traffic data from working CCS subnetworks", IEEE JSAC, 1994.

....understood perfectly. This is partly due to uncertainties in the traffic characteristics, and to the difficulties in characterizing the traffic arrival models. Analysis of traffic data from networks and services such as ISDN traffic [10] Ethernet LAN s [8] Common Channel Signaling Network (CCNS) [15], and Variable Bit Rate (VBR) video [13] have all convincingly demonstrated the presence of features such as self similarity, long range dependence, slowly decaying variances, heavy tailed distributions and fractal dimensions which are mostly characteristics of fractal processes. Leland and Wilson ....

D.E.Duffy, W.Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks", IEEE JSAC, 1994.

....timescale fluctuations in network traffic. Keywords Wide Area TCP traffic, Multi fractal Scaling, Performance Analysis I. INTRODUCTION It is now generally accepted that sufficiently aggregated packet network traffic exhibits self similar scaling over a wide range of timescales [1] 2] 3] [4], 5] The performance and traffic engineering implications of this property have been extensively explored (see for example [6] 7] and a self similar traffic model, Fractional Brownian Motion (FBM) has been proposed as a parsimonious and tractable model of packet traffic [8] FBM is an ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein and W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working Subnetworks", IEEE J. Selected Areas in Comm., Vol. 12, No. 3, pp. 544-551, 1994.

....timescale fluctuations in network traffic. Keywords Wide Area TCP traffic, Multi fractal Scaling, Performance Analysis I. INTRODUCTION It is now generally accepted that sufficiently aggregated packet network traffic exhibits self similar scaling over a wide range of timescales [1] 2] 3] [4], 5] The performance and traffic engineering implications of this property have been extensively explored (see for example [6] 7] and a self similar traffic model, Fractional Brownian Motion (FBM) has been proposed as a parsimonious and tractable model of packet traffic [8] FBM is an ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein and W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working Subnetworks ", IEEE J. Selected Areas in Comm., Vol. 12, No. 3, pp. 544551, 1994.

....characterized by Markov chains, and second, receivers require rigid delay bound. Recent studies show that network traffic exhibits long range dependence, a phenomenon not consistent with the first assumption. Long range dependence has been observed in isdn traffic [MH 91] telephone traffic [DMRW94] local area ethernet traffic [LTWW94] wide area Internet traffic [PF94, KM94] worldwide web traffic [CB96] video traffic [BSTW95, GW94] and audio traffic [Flo96a] Long range dependent traffic has been shown to effect queue behavior and loss rate [LW91, ENW96, GB96] and its cause has been ....

.... H12, H14, and H26 H15 H11 and H17 H2 H10 and H25 H16 H5 and H9 H3 H4 and H19 H17 H12 H4 H18 H18 H5, H6, and H11 H5 H14 and H25 H19 H5 H6 H18 H20 H5 H7 H17 H21 H9 H8 H4, H5, H26 H22 H6 H9 H3 and H19 H24 H12 and H17 H10 H3 and H18 H25 H6 and H14 H12 H4 H26 H9 and H14 H13 H17 H27 H4 studies ( LTWW94, DMRW94, PF94, KM94, GW94, BSTW95] have shown that network traffic often exhibits long range dependence (lrd) with the implications that congested periods can be quite long and a slight increase in the number of active connections can result in large increase in packet loss rate [PF94] The authors of ....

[Article contains additional citation context not shown here]

D.E. Duffy, A.A. McIntosh, M. Rosenstein, and W. Willinger. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks ". IEEE Journal of Selected Areas in Communication, 12(3):544-- 551, Apr. 1994.

....burstiness, network traffic dynamics is also effected by the arrival pattern and duration of flows. We use exponentially distributed lifetimes with the Markov ON OFF source model, following[14] The duration of Pareto ON OFF sources, however, are taken from a lognormal distribution, following [3, 5]. The interarrival times of all flows are exponentially distributed [16] Table 1: Six Instantiations of the Two Source Models Model Parameters TB Filter Switch Parameters Model Name p pkt I N p=a r tkn b max D D Upsilon S b p pkt sec msec pkts sec tkns qlen msec msec ( ptime sec EXP1 ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, and W. Willinger. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks". IEEE Journal of Selected Areas in Communication, 12(3):544--551, Apr. 1994.

....follows the shape until the 57 seconds, it misses the final rise due to the time limit. Note that because of the property of this time limit, the histogram of the PAMR system does not reflect the heavy tailed nature of the call holding time distributions observed by Bolotin [9] and Duffy et al. [11] in telephone networks. 0 20 40 60 80 100 0 120 s 0 10 20 30 40 Tr(s) System B System A Figure 1. Mean remaining time T r vs. elapsed message holding time. As we mentioned earlier, a important property of the exponential distribution is its lack of memory. As can be seen in Figure 1, the mean ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks", IEEE Journal on Selected Areas in Comm., Vol. 12, pp. 544-551, April 1994.

.... using traditional packet traffic models but, instead, seems to be best described with the 6 help of heavy tailed distributions(Zeta and Pareto distributions) 16] In CCSN SS7, the call holding time distribution for calls originating during both high and light traffic periods is heavytailed [17]. The distribution of bytes per FTPDATA burst is heavy tailed[4] and TELNET connection size in packets has a long tailed distribution[18] The aggregated traffic arriving at each input port is approximated by a Poisson Zeta model PZ a,r ( The numbers of bursts arriving at any time t . ....

Duffy, D E, McIntosh, A A, Rosentein, M and Willinger, W "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks," IEEE J. Select. Areas Commun., Vol.12, No.3 (March 1994), pp. 544-551

....The main conclusion is that mixtures of lognormal distributions suit the channel holding time of most of the services supplied by telephone systems (individual voice, subscriber line group, fax, etc. Bolotin found squared coefficients of variation between 3.2 and 4.8. Duffy et al. show in [8] how the channel holding time distribution of some telephone systems can be of the heavy tailed type (decays as a negative power of the time) and the coefficient of variation extremely large. The exponential distribution is not heavy tailed (decays faster) while the lognormal distribution is. ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks", IEEE Journal on Selected Areas in Communications, Vol.12, num. 3, pp. 544-551, Apr. 1994.

....as a slower decay of the autocorrelation function (as implied by (3.1) 3. 3 Implications of self similarity in B ISDN networks Evidence to suggest that self similarity does occur in teletraffic scenarios exists for Ethernet LAN and WAN [33] 48] Common Channel Signalling Network (CCSN SS7) [18], Integrated Services Digital Network (ISDN) 49] and VBR video [50] Although counter claims to these results have been made, 17] in this section we wish to consider the implications of self similarity on network performance. If self similarity does exist in broadband teletraffic then [20] ....

D. E. Duffy, A. A. McIntosh, M. Rosenstein, and W. Willinger, "Statistical analysis of CCSN/SS7 traffic data from working subnetworks," IEEE Journal on Selected Areas in Communications, vol. 12, no. 3, pp. 544--551, 1994.

....All of them are ON OFF processes. They differ in the distribution of their ON time and call holding time (CHT, which we will also call flow duration or flow lifetime ) One of these is the rather standard two state Markov process used widely in the literature. Recent studies ( LTWW94, DMRW94, PF94, KM94, GW94, BSTW95] have shown that network traffic often exhibits long range dependence (LRD) with the implications that congested periods can be quite long and a slight increase in number of active connections can result in large increase in packet loss rate. Reference [PF94] further ....

.... rate C C Host Network Packets transmitted at rate G, r = G = C Q packet data queue averaging interval r = N (N M) a =rp R = p On Off Traffic generator Figure 3: ON OFF traffic model with token bucket filter the LRD models, however, are taken from a lognormal distribution, following [Bol94, DMRW94] The interarrival times of all flows are exponentially distributed [PF94] 5 4.3 Parameter Choices We chose six instantiations of the above three source models, as summarized in Table 1. In the table, p = 1 means that after each OFF time, packets for the next ON period are transmitted back to ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, and W. Willinger. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks". IEEE Journal of Selected Areas in Communication, 12(3):544--551, Apr. 1994.

....are given to demonstrate our results. 1 Introduction In the past 5 years, large amounts of traffic measurements from working packet networks (including Ethernet LANs, wide area TCP IP, CCSN SS7, ISDN, VBR video, Frame Relay, and ATM) have been collected and analyzed. The results reported in [1, 3, 4, 10, 11, 13, 16, 17, 18, 21] have been striking for two reasons: i) these studies demonstrate that it is possible to distinguish clearly between actual packetnetwork traffic and traffic generated by traditional Markovian models, and (ii) in sharp contrast to the traditional packet traffic models, aggregate packet streams ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein and W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working Subnetworks," IEEE JSAC , Vol. 12, No. 3, pp. 544-551, 1994.

....fractal in nature. Other studies have shown that self similarity is a feature Variable BitRate Video Traffic [BSTW94] Work done at Lawrence Berkeley Labs [PF94] shows that certain major components of internet traffic exhibit self similarity. Studies of Common Channel Signaling Networks (CCSNs) [DMRW94], and Integrated Services Digital Networks (IDSN) GW94] have also uncovered self similarity. 1.1 An Intuitive Description of Self Similarity Intuitively, self similar phenomena display structural similarities across a wide range of time scales. This intuition is illustrated graphically in Fig. ....

D. E. Duffy, A. A. McIntosh, M. Rosenstein, W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working Subnetworks," IEEE Journal of Selected Areas in Communications 12, pp.544-551, 1994.

....model in our simulations. All of them are ON OFF processes. They differ in the distribution of their ON time and call holding time (CHT, which we will also call flow duration or flow lifetime ) One of these is the twostate Markov process used widely in the literature. Recent studies ( LTWW94, DMRW94, PF94, KM94, GW94, BSTW95] have shown that network traffic often exhibits long range dependence (LRD) with the implications that congested periods can be quite long and a slight increase in number of active connections can result in large increase in packet loss rate [PF94] Reference [PF94] ....

....and the mix of transport protocol, source model, token bucket filter, and service request ascribed to each flow. We ascribe exponentially distributed lifetimes to the EXP model, following [Mol27] The duration of for LRD sources, however, are taken from a lognormal distribution, following [Bol94, DMRW94] The interarrival times of all flows are exponentially distributed [PF94] 4.3 Parameter Choices We chose six instantiations of the above three source models, as summarized in Table 2. In the table, p = 1 means that after each OFF time, packets for the next ON period are transmitted back to ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein, and W. Willinger. "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks". IEEE Journal of Selected Areas in Communication, 12(3):544--551, Apr. 1994.

....to the modeling of broadband ISDN and ATM networked systems. Leland et.al. at Bellcore Morristown Research Center have analyzed millions of packets on several Ethernet LAN s and millions of frame data by Variable Bit Rate (VBR) video services [11, 12] In [11, 12] and in numerous other studies [1, 2, 3, 4, 8, 10] network packet traffic is characterized by high variability or burstiness over a wide range of time scales. In other words, the network traffic looks the same when measured over time intervals ranging from milliseconds to minutes to hours, etc. 5] Data traffic of this type is said to be ....

D. E. Duffy, A. A. McIntosh, M. Rosenstein, and W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working Subnetworks," IEEE J. Select Areas Commun.,, vol. 1, no. 3, pp. 544-551, 1994.

.... cities and extinction events [K93, M63] Leland and Ott also found that a Pareto distribution with 1:05 fi 1:25 is a good model for the amount of CPU time consumed by an arbitrary process [LO86] In communications, heavy tailed distributions have been used to model telephone call holding times [DMRW94] and frame sizes for variable bit rate video [GW94] The discrete Pareto (Zipf) distribution [A83, p.95] P [x = n] 1= n 1) n 2) for n 0: arises in connection with platoon lengths for cars at different speeds traveling on an infinite road with no passing [A83, p.95] F66, p.40] a model ....

....network traffic. Following [LTWW94] we define a distribution as heavy tailed if: P [X x] cx 0fi ; as x 1; fi 0: 1) By this, we mean that for some fi and some constant c, the ratio P [X x] cx 0fi ) tends to 1 as x 1. This definition includes the Pareto and Weibull distributions [DMRW94]. A more general definition of heavy tailed defines a distribution as heavy tailed if the conditional mean exceedance (CME x ) of the random variable X is an increasing function of x [HK80] where CME x = E[X 0 xjX x] Using this second definition of heavy tailed, consider a random variable X ....

D. Duffy, A. McIntosh, M. Rosenstein, and W. Willinger, "Statistical Analysis of CCSN/SS7Traffic Data from Working CCS Subnetworks," IEEE JSAC, 12(3), pp. 544551, April, 1994.

....insights into the traffic management of high speed networks. 1 Introduction In the past 2 3 years, large amounts of traffic measurements from working packet networks (including Ethernet LANs, WANs, CCSN SS7, ISDN, and VBR video over ATM) have been collected and analyzed. The results reported in [1, 5, 7, 13, 25, 26, 40, 23, 33, 35, 36] have been striking for two reasons: 1) these studies demonstrate that it is possible to clearly distinguish between actual packet network traffic and traffic generated by widely employed theoretical models, and (2) in sharp contrast to the traditional packet traffic models, aggregate packet ....

....measurement studies. Prominent among these recent studies are the statistical analyses of (i) Bellcore s Ethernet LAN traffic measurements [25, 26] ii) WAN traffic traces collected at Berkeley [35, 36] iii) traffic measurements collected from working Common Channel Signaling (CCS) subnetworks [5], iv) packet traces collected in an ISDN office automation environment [33] and (v) VBR video traces [13, 1, 20] Note that because of the use of highestquality traffic monitoring equipment and because of current data storage capabilities that are practically unlimited, the resulting sets of ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein and W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working Subnetworks", IEEE J. Select. Areas in Communications, Vol. 12, No. 3, pp. 544-551, 1994.

....canvas (the two autocorrelation curves are visually indistinguishable) Such source models can be used to generate synthetic streams of realistic traffic to drive simulations of communications networks. 2. 6 SELF SIMILAR TRAFFIC MODELS Recent traffic studies from working packet networks [7, 24, 33, 59, 97, 75] have revealed new features of packet traffic that have gone unnoticed in the traffic modeling literature, yet seem to have serious implications for designing, engineering and controlling future high speed networks. The measured data demonstrate convincingly that packet traffic can be ....

Duffy, D.E., McIntosh, A.A., Rosenstein, M. and Willinger, W., "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks", IEEE J. Select Areas Communications 12:3 (1994), 544--551.

....in the Terabyte range have been collected by Bellcore researchers from the Common Channel Signaling Network (CCSN) that relies on the Signaling System Number 7 protocol (SS7) for communication. Small segments of this enormous data set have been analyzed in Duffy, McIntosh, Rosenstein and Willinger [9] (see also Bolotin [4] CCSN traffic measurements were also collected from the British Telecom (BT) network [1] Cinotti, Dalle Mese, Giordano and Russo [5] report on a traffic collection and analysis effort involving a cluster of LANs and a Metropolitan Area Network (MAN) corresponding to a ....

D.E. Duffy, A.A. McIntosh, M. Rosenstein and W. Willinger, "Statistical Analysis of CCSN/SS7 Traffic Data from Working CCS Subnetworks", IEEE Journal on Selected Areas in Communications 12, pp. 544-551, 1994.

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