01 May 2012 19:30
In prior experiments, we found page load times to be highly variable over real 3G and 4G cellular networks, making it hard to draw conclusions without running hundreds of experiments per site in order to estimate the statistical distribution. To reduce this variability, the phone was tethered to the desktop machine hosting the server using a USB connection, and traffic shaping was applied to the tethered connection using Dummynet. We emulated a 3G network with uplink bandwidth of 1 Mbps, downlink bandwidth of 2 Mbps, and a round-trip delay of 150 ms. These values were chosen as representative of cellular network performance in the United States. Note that packet loss was not included in the traffic shaping parameters, since cellular networks hide packet loss at the PHY layer, and our previous experiments have shown a TCP-level packet loss of less than 1% over typical cellular networks.
Google mesure la performance de SPDY sur les réseaux mobiles... en simulant le réseau mobile. Ils auraient aussi pu simuler SPDY, comme ça on aurait eu une couverture méthodologique à 100%.