Ferramenta de aferição periódica e automatizada de métricas da qualidade de serviço da conexão residencial à internet

In Brazil, the Anatel Agency specifies the minimum Quality of Service (QoS) that internet service providers (ISPs) should maintain. Although there may be potential interferences in internet quality measurements on popular tools such as Speedtest.net, when the user tests against a Speedtest server hosted within the ISP infrastructure, they are actually evaluating the quality of access to their intranet, rather than the actual quality of internet access they are paying for. Furthermore, conventional benchmarking tools are not periodic, limiting the network assessment to a specific instant (without capturing the average connection quality), and it also has the limitations of the client-server model. As a solution to this problem, in this work a peer-to-peer (P2P) measurement tool named Peertest has been developed for periodic measuring of metrics related to the quality of service maintained by the ISP in a typical residential internet access connection. The BitTorrent DHT network was leveraged as a rendezvous point for Peertest peers and STUN followed by the UDP Hole Punch technique aimed to achieve end-to-end connection between them, even if they were behind NATs, so they could measure one against the other. The developed tool was able to measure latency and packet loss metrics in a more realistic manner, reporting a circadian rhythm and more heterogeneous values in comparison with Speedtest.net tests. The prototype was validated in a relevant operating environment (peer at home and peer on Amazon AWS). In the field tests, the download and upload speeds were limited due to some specific limitations in the experiment environment. This was likely due to traffic shaping applied by some ISPs. In addition to the developed tool, the utilization of Transport Protocol Translation and UDP Hole Punch techniques was documented, along with several peculiarities observed in ISPs in order to achieve end-to-end connection. Furthermore, a pair selection algorithm that is capable of accommodating both similar and derivative solutions is also presented. The tool can be obtained directly via a Docker container image. The tool’s source code and the results of the experiment are available on GitHub.