Scholarlink

<p>The problem of modeling the end-to-end delay in computer networks without Kleinrock’s independence assumption (KIA) has not been solved since 1961. Computer networks are modeled with the following assumptions: packet arrivals are Poisson, packets are routed based on a fixed routing strategy, packet lengths are exponentially distributed and remain unchanged when they traverse from node to node in networks. In this paper, we first introduce a distribution <i>C</i>(<b>p, θ</b>), which generalizes the hypoexponential distribution. Based on this distribution, we develop a computationally simple method to model three end-to-end delay measures, namely the probability distribution functions, the average delay and the jitter of end-to-end delay. We then show that this method provides a “good” prediction about the above delay measures. This is done by simulating a 40-node random network with 1560 packet flows under two network loads: light and high. For those 1560 flows under two network loads, the end-to-end delay distribution functions using our method generally pass the Kolmogorov-Smirnov test at a 5% significance level; the relative errors in our predicted average end-to-end delay and jitter values are all less than 0.066.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

End-to-end delay model in computer networks without Kleinrock’s independence assumption

  • 期刊论文

  • 发表日期：

    2026年6月16日

查看全文
  • Yu Chen,
  • Bowen Xu,
  • Qimei Cui,
  • Xiaofeng Tao,
  • Ping Zhang

摘要

The problem of modeling the end-to-end delay in computer networks without Kleinrock’s independence assumption (KIA) has not been solved since 1961. Computer networks are modeled with the following assumptions: packet arrivals are Poisson, packets are routed based on a fixed routing strategy, packet lengths are exponentially distributed and remain unchanged when they traverse from node to node in networks. In this paper, we first introduce a distribution C(p, θ), which generalizes the hypoexponential distribution. Based on this distribution, we develop a computationally simple method to model three end-to-end delay measures, namely the probability distribution functions, the average delay and the jitter of end-to-end delay. We then show that this method provides a “good” prediction about the above delay measures. This is done by simulating a 40-node random network with 1560 packet flows under two network loads: light and high. For those 1560 flows under two network loads, the end-to-end delay distribution functions using our method generally pass the Kolmogorov-Smirnov test at a 5% significance level; the relative errors in our predicted average end-to-end delay and jitter values are all less than 0.066.