2. TRAFFIC CAPTURE AND STATISTICS
For an evaluation of the traffic we performed several measurements capturing the Ethernet traffic at the Office Communication server. The capturing of the traffic was performed using the popular Wireshark tool [7], which reports complete
Ethernet frame sizes, including the Ethernet overhead per captured frame of 14 bytes. We initially dropped frames that do not adhere to standard (with sizes greater than 1514 bytes, as the network did not allow Jumbo frames) and aggregated the remaining captured frames on a per-second basis to determine the bandwidth. Specifically, we aggregate the size of all Ether- net frames within a specific second from the beginning of the measurement and denote the resulting bandwidth as Xt [bps]. We denote the average bandwidth as X [bps] and Xmax [bps] as the maximum bandwidth observed. Subsequently, we denote the Peak-to-Mean ratio (as one measure for the variability of
the underlying traffic) of the bandwidth defined as P tMX .
2 We calculate the variance SX and standard deviation SX of the observed bandwidth usages to calculate the coefficient of variation, commonly used to measure the variability of an underlying variable, and denote it as CoVX .
The Hurst parameter H, is one key measure of selfsimilarity [8], [9] for an underlying process and commonly used to estimate the long-range dependence of a stochastic process. A Hurst parameter of H = 0.5 indicates absence of self-similarity whereas H = 1 indicates long-range dependence. We additionally note that while the Hurst parameter is mathematically well-defined, its determination is not clearly defined, but commonly performed through estimations, see,e.g., [10]. We employ the R/S statistic [11]–[13] to estimate the Hurst parameter for the captured traffic. Specifically, we estimate the Hurst parameter H as the slope of a log-log plot (also referred to as pox plot or adjusted scaled range plot) of the R/S statistic using a least squares fit. In addition, we calculate the Hurst parameter using the technique outlined in [14] as an additional metric. For a more visual approach to the identification of network bandwidth usage and its self-
similarity, we utilize the autocorrelation function, see, e.g.,[15].improvement
2. TRAFFIC CAPTURE AND STATISTICS
For an evaluation of the traffic we performed several measurements capturing the Ethernet traffic at the Office Communication server. The capturing of the traffic was performed using the popular Wireshark tool [7], which reports complete
Ethernet frame sizes, including the Ethernet overhead per captured frame of 14 bytes. We initially dropped frames that do not adhere to standard (with sizes greater than 1514 bytes, as the network did not allow Jumbo frames) and aggregated the remaining captured frames on a per-second basis to determine the bandwidth. Specifically, we aggregate the size of all Ether- net frames within a specific second from the beginning of the measurement and denote the resulting bandwidth as Xt [bps]. We denote the average bandwidth as X [bps] and Xmax [bps] as the maximum bandwidth observed. Subsequently, we denote the Peak-to-Mean ratio (as one measure for the variability of
the underlying traffic) of the bandwidth defined as P tMX .
2 We calculate the variance SX and standard deviation SX of the observed bandwidth usages to calculate the coefficient of variation, commonly used to measure the variability of an underlying variable, and denote it as CoVX .
The Hurst parameter H, is one key measure of selfsimilarity [8], [9] for an underlying process and commonly used to estimate the long-range dependence of a stochastic process. A Hurst parameter of H = 0.5 indicates absence of self-similarity whereas H = 1 indicates long-range dependence. We additionally note that while the Hurst parameter is mathematically well-defined, its determination is not clearly defined, but commonly performed through estimations, see,e.g., [10]. We employ the R/S statistic [11]–[13] to estimate the Hurst parameter for the captured traffic. Specifically, we estimate the Hurst parameter H as the slope of a log-log plot (also referred to as pox plot or adjusted scaled range plot) of the R/S statistic using a least squares fit. In addition, we calculate the Hurst parameter using the technique outlined in [14] as an additional metric. For a more visual approach to the identification of network bandwidth usage and its self-
similarity, we utilize the autocorrelation function, see, e.g.,[15].improvement
การแปล กรุณารอสักครู่..
