Nexus One uses. Since that mechanism tears down the RRC connection in between each video content burst, the connection must be re-established every time a new burst starts, which requires a lot more signaling than transition from CELL PCH to CELL DCH state. For the same reason, reshaping the traffic into bursts while not having the CELL PCH state enabled increases the signaling load more than in the case where that state is enabled. The reason why signaling load is even higher in the case of Nexus One compared to the case where CELL PCH state was disabled for N900 and Lumia is the YouTube background traffic, In the case of Nexus One, those periodic packets emerge right in the middle of the video bursts which causes extra state transitions each time. The shorter timer case for N900 shows a surprisingly high increase in signaling load when the proxy is applied. The reason turned out to be that N900 applied legacy FD after some bursts. The logic explaining how N900 decides when to apply that mechanism remains unclear to us. We also investigated whether introducing the proxy increases the amount of retransmitted bytes. The rationale is that when a burst arrives to the radio access network, the device needs to transition to CELL DCH state which causes non-negligible delay (in the order of a few seconds in the worst case) during which TCP retransmission timeout may expire for the first packets which will then be retransmitted. Indeed the case. For N900 and Nexus One, which download the video content slower than Lumia, the ratio of retransmitted bytes systematically increased from almost zero to somewhere between 1-2%. Lumia’s fast download already caused a larger amount of bytes to be retransmitted and, therefore, the increase was not as substantial.