When designing a cross-border architecture and its corresponding roaming mechanisms, i.e., choosing between Home Routing Roaming (HRR) and Local Breakout (LBO), one should consider the specificity of the services deployed and their requirements in terms of throughput, latency, and interruption time.  

In HRR (depicted in the figure below), the user data is forwarded from the visited User Plane Function (v-UPF) to the home UPF (h-UPF) before reaching the Data Network (DN), which leads to additional latency corresponding to the time required to send the data from v-UPF to h-UPF. In contrast, in LBO the data is sent directly to the DN through the v-UPF without passing through the h-UPF as shown in the figure, and therefore reducing the latency. However, when a User Equipment (UE) crosses the border using LBO, it might experience longer interruption time as all core functions have to be moved to the visited network and new IP address has to be assigned to the UE. 

Some services, such cooperative perception, are very sensitive to interruption time as the V2X messages will not be received during this period. For this type of service, HRR can be more suitable, especially if no roaming optimization is deployed, as it induces less interruption time than LBO. Other services, such as tele-operated driving, require low latency as the reaction to any incident should be in real time. For this type of service, LBO is more suitable than HRR. It should be noted that the interruption time happens only when crossing the border, which is a very well-defined location. To overcome the longer interruption time QoS prediction method can be used and the speed of the tele-operated vehicle can be reduced (or even the vehicle can be stopped) at the border until the roaming is established. However, longer latency in all the journey inside the visited country will mean that the tele-operated vehicle must be driven always with low speed. In addition, real-time services, such as infotainment, the application server should follow the UE by connecting the latter to the closest MEC. For this type of services, HRR does not make sense as the data will always be forwarded to the home UPF; as it can be seen from the above figure user data is always forwarded to the DN in the home network even when the UE is in the visited network. 

As we have seen, these two roaming mechanisms provide different performance results and their usage is tightly related to the requirements of the services used in the network. 5GMED will explore several roaming optimization techniques tailored to the services presented in the different use cases of the project.