Synthèse
To understand how traffic changes throughout a match, we had to establish what it is “normally.” One way to do this is by looking at raw request volumes, or the amount of traffic we see on our network per country. But these amounts vary per country (the amount of daily traffic in the United States is always a larger number than the traffic in Portugal), which makes it difficult to establish a globally applicable baseline. Instead, we defined "normal" using the median traffic of the four precedin
To understand how traffic changes throughout a match, we had to establish what it is “normally. ” One way to do this is by looking at raw request volumes, or the amount of traffic we see on our network per country. But these amounts vary per country (the amount of daily traffic in the United States is always a larger number than the traffic in Portugal), which makes it difficult to establish a globally applicable baseline. Instead, we defined "normal" using the median traffic of the four preceding weeks: a month-long window that provided a stable, per-minute reference and smoothed out day-to-day noise. We also wanted to know whether traffic rose or fell relative to that baseline, but a plain difference wouldn't let us compare a high-volume country against a low-volume one.
Instead, we used the ratio of current to baseline traffic, expressed as a log₂ value: the log makes increases and decreases symmetric around zero (+1 = twice normal, −1 = half). In other words, a score of zero means traffic is perfectly normal, a positive number shows a spike, and a negative number shows a drop.