Hi Chanka,

When we analyze latency, we normally compute the average and percentiles. However, these descriptive statistical measures do not give us an understanding of the overall behaviour of latencies. To get a better understanding of how the latencies behave we need to look at the latency distribution. This can be done by 1) constructing a histogram (a simple method) or 2) by fitting a distribution to the data. Here I have done both.

The relative frequency histograms (in this article) shows the probability that the latency (i.e. random variable) is in a given latency range (where the range = bin size). The area under a relative frequency histogram = 1.

When constructing histograms there is an optimal way to compute the bin size (to accurately capture the behaviour). I notice that in the histogram plots, I haven’t specified the x values of individual bins (but you can sort of get an idea by looking at values in the x axis)

Note that to get the probability that the latency in specific latency range, say from x2 to x2, from the probability density function (i.e. f(x)), we have to integrate f(x) from x1 to x2 with respective to x.