The IoT slice obtains its bandwidth from the WebData slice, with a maximum of 60 Mbps, without affecting the Control and Streaming slices. The Streaming slice obtains its bandwidth from the IoT and WebData slices, at a maximum of 80 Mbps, but it does not affect the bandwidth of the Control slice. Thus, the Control slice reaches the saturation limits as the channel in 100 Mbps, once it obtains bandwidth from other slices. It can be seen that, when each slice requires more bandwidth, it obtains the bandwidth from lower priority slices. This figure clearly shows the behavior of traffic prioritization in the different slices. In this way, the traffic sent by Iperf can be treated as Control, Streaming, IoT and WebData traffic.įinally, Figure 8 shows the bandwidth obtained in the third scenario, where a single UE is connected to the same four slices used in the second scenario, i.e., Control, Streaming, IoT and WebData. With these rules, when the Iperf tool sends packets to the same ports used by the already-characterized traffic, these data flows are redirected to the previously mapped slices. Subsequently, a series of redirection rules are added, only for this saturation test, due to the fact that the saturation of a network with real IoT traffic is not easy because they have a small packet size. In this way, the proposed approaches characterize the traffic and map it to the corresponding slices. In order to use it to achieve network saturation and to obtain metrics, each traffic is first generated using the aforementioned tools (Ping, FFmpeg, MQTT and Wget). This tool was selected because it is a popular command-line tool used to diagnose network problems by measuring the maximum network throughput. To carry out the measurement of these metrics, the active measurement tool Iperf was used to obtain the maximum achievable bandwidth in IP networks. Since in our scenario this collection time is not critical for the correct operation of the slicing selection algorithms, it was established that the traffic classifier would collect traffic at an interval of 10 s, to ensure high efficacy and accuracy in the decisions. In our tests, carried out for the five types of traffic defined in Table 5, it was concluded that using 1 s as the window of time for the evaluation algorithm (the minimum collection interval allowed by the nDPI traffic classifier) was enough to accurately detect traffic, but depending on the protocol/application, more time may be necessary for correct detection. According to the literature, there is a rule stating that only eight packets per direction allow the nDPI algorithm to accurately identify the protocol or application. In the case of the nDPI traffic classifier, a single packet is usually necessary to determine most User Datagram Protocol (UDP)-based protocols widely employed in the industry and in automotive applications. The time interval is dependent on the application or protocol to be detected. Moreover, to obtain a proper and efficient traffic classification, it is important to determine the volume of packets to collect from the network to conclude an accurate classification while also quickly adapting to traffic-type shifts.
0 Comments
Leave a Reply. |