Self-similarity of Internet packet delay

Borella, Michael S.; Uludag, Suleyman; Brewster, G.B.; Sidhu, Ikhlaq

doi:10.1109/icc.1997.605360

Cited by 41 publications

(22 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The parameters of the Gamma distribution have been observed to depend upon the path (e.g., regional, backbone) and the time of the day. The heavy tail behavior is due to the presence of self-similarity in Internet packet delay [3]. Typical queueing models like M/M/1, M/G/1 and using Fractional Brownian Models (fBm) for traffic models have been shown to underestimate average queueing delays for link utilization below 70% [18].…”

Section: Introductionmentioning

confidence: 99%

On Kleinrock’s Independence Assumption

Popescu

Constantinescu²

2011

Network Performance Engineering

View full text Add to dashboard Cite

Abstract. This chapter is concerned with the impact of traffic correlations on the end-to-end delay of packets in a chain of IP routers represented by an open tandem queueing system. Particular consideration is given to Kleinrock's independence assumption on the performance impact of traffic correlation on packet switched networks with Poisson arrival processes and exponential packet length distributions. According to this assumption, any traffic correlations can be ignored and the effect on delay performance is negligible, subject to sufficient traffic mixing and moderate-to-heavy traffic loads. In this context, results associated with measurements, traffic modeling and delay analysis of an actual chain of IP routers are reported from experiments conducted at the Blekinge Institute of Technology in Karlskrona, Sweden. It is shown that no experimental evidence was found in support of Kleinrock's independence assumption as traffic correlation has an adverse effect on the end-to-end delay of the tandem queueing system.

show abstract

Section: Introductionmentioning

confidence: 99%

On Kleinrock’s Independence Assumption

Popescu

Constantinescu²

2011

Network Performance Engineering

View full text Add to dashboard Cite

show abstract

“…If delay traces span over hours or days it may raise a question whether or not the LRD-like effect revealed by certain statis- 3 In this study we call the log 2 E j vs. j plot as the energy-scale plot. tical techniques is actually caused by non-stationarity in the delay process.…”

Section: Robustness Of the Wavelet-based Analysesmentioning

confidence: 99%

“…If the linear relationship between log 2 E j and scale j hold for large time scales, it suggests the incremental process of an asymptotically self-similar process. Performing a least-squares fit on the linear portion of the energy-scale plot 3 , we can estimate the Hurst parameter H, so the scaling analysis also serves as an H estimator.…”

Section: Discrete Wavelet Transform and Scaling Analysismentioning

confidence: 99%

“…Our research is closely related to the studies carried out in [3,10]. The study in [3] first explored the long-range dependence (LRD) characteristics of the measured round-trip delay time series.…”

Section: Introductionmentioning

confidence: 99%

“…The study in [3] first explored the long-range dependence (LRD) characteristics of the measured round-trip delay time series. The study in [10] also found the LRD characteristics of the round-trip delay time series measured by NTP.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Investigating the scaling behavior, crossover and anti-persistence of Internet packet delay dynamics

Mills²

Seamless Interconnection for Universal Services. Global Telecommunications Conference. GLOBECOM'99. (Cat. No.99CH37042)

View full text Add to dashboard Cite

Relying on the wavelet-based time-scale analysis techniques and a so called height difference correlation function method, we studied the dynamics of the delay traces measured in the Internet with sampling interval ranging from 20 ms to 1 min. Scaling analyses showed that the delay traces have a complicated scaling behavior changing with both the time scales and the sampling intervals. A significant observation of our research was that the nature of the delay traces changes from a LRD noise-like time series (e.g., fractional Brownian noise) to a self-similar one with an outer cutoff scale when the sampling intervals decrease (we refer this as crossover). It turns out that the delay jitter time series derived from a sampled delay trace with small sampling interval is LRD noise with a Hurst parameter H 0:5 (anti-persistence), indicating that the current Internet works in a controlled state. We also used a wavelet-based partition function method to investigate the possible multifractality of the delay traces. We analyzed the underlying physical mechanisms of the observed scaling behavior, crossover and anti-persistence, and discussed the implications of our findings on several aspects of networking engineering.

show abstract

Performance evaluation of TCP session considering long‐range dependence of RTT

Hagiwara

Matsuda

Yamamoto

2004

Electron. Comm. Jpn. Pt. I

View full text Add to dashboard Cite

SUMMARYRecent research has shown that long-range dependence is found not only in network traffic volume, but also in the propagation delay time of packets. The round-trip propagation time of packets, which is known as the roundtrip time (RTT), is one of the most important parameters for determining TCP performance. The ability to understand its characteristics can contribute to an improvement in TCP performance. Therefore, in this paper, to determine the impact of the long-range dependence of the RTT on TCP performance, the authors looked at the two congestion control phases known as congestion avoidance and slow start in TCP window control and used computer simulations to evaluate performance. First, the evaluation related to the impact on the congestion avoidance phase showed that higher long-range dependence and larger transfer files caused greater variability in the file transfer completion time. Next, the evaluation related to the impact on the slow start phase, which evaluated the impact of the long-range dependence of the RTT on the TCP retransmission timeout (RTO), showed that higher long-range dependence causes unnecessary timeouts, which end up causing the resending of packets that have not actually been lost, to occur more frequently. The authors also performed investigations for mitigating the impact of the long-range dependence of the RTT and made a number of modifications to the existing RTO estimation algorithm to show that TCP performance can be improved.

show abstract

Self-similarity of Internet packet delay

Cited by 41 publications

References 17 publications

On Kleinrock’s Independence Assumption

On Kleinrock’s Independence Assumption

Investigating the scaling behavior, crossover and anti-persistence of Internet packet delay dynamics

Performance evaluation of TCP session considering long‐range dependence of RTT

Contact Info

Product

Resources

About