Is the burst length periodicity a result of sever using different periodic processes that are out of phase wrt each others?

We can start to form a hypothesis as to what might cause such periodic behavior. The hypothesis is that there are a set of processes that are sampling the information about the arrivale rate of packets at the client and sending this information back to the sevrer. Then based upon this information the server will adjust its sending rate. Since we would like to sample this information at different times, so that we can learn about the delays that packets are experiencing early in the arrival of a burst of packets and later - we will assume that there are at least two sampling processes that are working out of phase with each other.

II2202-presenting-data-Maguire-20150920a_img_65.jpg
Plot of the burst lengths

Showing only the first 330 bursts

Transcript

So could we explain that? Well, if I remove the lines that are connecting them.