Throughput observed per second

Upload

Download

Statistical analysis of inter packet delay per measurement session

The measurement captures the delay between two consecutive packets and calculates the distribution of the delay in 0.1ms bin between 0 to 20ms. It also calculates the first moment up to the fourth moment (average, variance, skewness and kurtosis_ of the packet variation delay for the measurement campaign.

The graphs above measure the average and median of the delay between two successive packets during a measurement session. It measures the typical latency between the client and server. The closer the median to the average, the more symetrical the distribution of pdv but not necessarily true. A population of 100 samples with 50 zeros and 50 ones will give an average of 0.5 whereas the median is zero (or one depending on formula). A change in values in time may indicates change of distribution or parameter. Use the standard deviation, skewness and kurtosis below to further understand what is happening.

The graph above measures the standard deviation of the inter packet delays during a measurement session. It starts to give us an idea on how dynamic the jitter could be during the measurement. High value indicates that the inter packet delay range are wider compared to a lower value. Higher delay range means the increased jitter. It does not inform whether the delay have increased or decreased or whether it is due to change of the outlier volume during the measurement. Use skewness and kurtosis below to further understand the situation.

Statistically, for a normal distribution, a Negative Skewness indicates a left leaning bell curve while a Positive Skewness indicates a right leaning bell curve. Zero Skewness indicates a perfectly balanced bell curve. The graph above measures the skewness of the delay between two successive packets for each measurement session. One point plot for each session. A lower skewness value than usual indicates a more narrow inter packet delay variation or better latency. A higher skewness indicates increased latency. Use the standard deviation indicator above and the kurtosis below to have a better idea whether the jitter has increased or not.

The graph above measures the Kurtosis of the delay between two successive packets for each measurement sessions. Kurtosis primarily measures the tail of a distribution while it can also be an indicator on the flatness of the distribution itself. A high number indicates higher number of outliers compared to a low number. For a normal distribution, the kurtosis value is equal to 3.

A higher kurtosis value than usual indicates a wider inter packet delay variation or worse latency consistency, with increased outliers. This means that the inter packets delay distribution is much flatter or the tails are more significant compared to a lower value.This suggest higher jitter as well as indirectly shows the higher variability of the jitter itself. A lower kurtosis indicates the latency is concentrated to a central value with narrow distribution and therefore suggest a lower jitter variability.

Sample of PDV before and after profile upgrade

Throughput deviations

Upload traffic

Download traffic