1-2hit |
Ryoichi KAWAHARA Keisuke ISHIBASHI Takuya ASAKA Katsunori ORI
We propose a method of IP traffic management where the TCP performance at a bottleneck link is estimated from monitored data about the behavior of the number of active flows versus link utilization, which are both easy to measure. This method is based on our findings that (i) TCP performance remains constant as long as the link utilization is below some threshold value, but becomes degraded when it exceeds this value and (ii) the number of active flows increases linearly with link utilization up to the same value, and the increase becomes nonlinear above it. Though this threshold may vary depending on traffic/network conditions, our method requires neither predetermination of a threshold on the basis of assumed traffic conditions nor direct measurement of TCP performance.
Takuya ASAKA Katsunori ORI Hiroshi YAMAMOTO
Measuring the duration of flow of a TCP connection in an end-to-end path is important in the management of network performance, and when this is done, an administrator can manage the quality of the networks using the α percentile of the distribution. We propose a method of estimating the distribution of flow duration in an end-to-end path through active measurement using a small degree of traffic. This method of estimation is based on traffic characteristics that are observed in measuring traffic in actual networks. It imposes little additional load on networks and the time in computation required to estimate the distribution is also short. The distribution to be estimated is assumed as a log-normal distribution, and the mean and variance of the distribution of a target file size is estimated by using results of active measurements. Means and variances in various file sizes are estimated through the linear relationships between these values (or their logarithms) and file size. We also provide numerical examples using actual traffic data.