Network performance: deaf dialogue between arcep and free mobile |igeneration

In its annual barometer of quality of mobile services published a few days ago, the arcep pushes free in the nettles.The operator displays the worst average rate in 5G with 31 Mbit/s against 142 Mbit/s at Orange, and it is good last in average flow 2G/3G/4G/4G.

Free reacted in two ways to this study: by congratulating itself to see its deployment strategy validated by ARCEP ... and pointing to the statement protocol used by the regulator, which gives it such bad results!

According to Free, the "disparities" between the download speeds measured by ARCEP and those noted by tools like Speedtest or NPERF are the result of use by the authority of TCP Cubic.It is a mono-connections protocol (Monothread) which measures the download time of a file and deduces the flow.In principle, the arcep test consists in using a single connection flow between the mobile and the server where the file to download is hosted.

To make it short, the Free Mobile network has been optimized for use with several connections (Multithread), which make many online services such as iCloud, Wetransfer, Netflix, YouTube ... or even "almost all websites including the onearcep, ”recalls, perfidious, the operator.The latter recommends the TCP BBR protocol, more recent (it dates from 2016, against 2008 for Cubic), which "uses bandwidth and round-trip time and is approaching the theoretical optimum".

ARCEP defended the technical choice of mono-connections, explaining that it is "majority use on the Internet": "Even when several applications are open to a mobile, in practice, most of the time, only one of the flows is used at a given moment, ”explains the regulator.Who admits that crowdsourcing applications like those mentioned by Free are based on multi-connections, which can "explain the differences between the flows measured by ARCEP and those from such applications".

In short, each camp remains camped on its positions and we are witnessing a dialogue of deaf.Free delivers the results of surveys with the BBR TCP on 50 addresses in Paris, and compared to those of ARCEP, they are more advantageous for the operator facing "operators 2, 3 and 4" (bet that the colors usedreflect those of their logos):

For Free, its flows are "similar or superior to those of competitors" when all servers use TCP BBR and/or the measurement app opens several TCP connections.The operator draws two conclusions: on the one hand, that in "real use", subscribers benefit from high flows in 4G and even more in 5G.And on the other hand, that the protocol used by arcep does not reflect these same real uses of mobile subscribers and the flows of free mobile subscribers.