Mobile Device Virtualization through State Transfer (MIGRATE) proposes to use a virtual representation of mobile devices to gather and process data at the edge of the infrastructure network, at the same time a proper management of the mobility of the nodes is considered. Virtual mobile devices, implemented as VxFs, follow the end device movement through a “migration” approach among supported virtualization domains.
Virtualization of computing and network functions is a key advance of 5G, and it can be used to offload constrained devices from processing and data management tasks. The direct instantiation of virtual substitutes for physical devices has been proved to be beneficial in terms of device access delay, reliability against wireless disconnections or data cache, as the SURROGATES proposal showed (https://5ginfire.eu/surrogate/). A way of exploiting Multi-access Edge Computing (MEC) capabilities of 5G is maintaining these virtual substitutes near or within the access network. However, a proper support of device mobility should be provided as it traverses access networks belonging to different administrative domains. In this case, virtual mobile nodes must be able to “migrate” to new edge virtualization domains.
Outperforming the work carried out in SURROGATES, the MIGRATE proposal generalizes the idea of virtual mobile devices (vMDs) and provides an efficient and novel solution for the transference of these software entities to follow the real location of mobile devices (MDs). This is carried out by using Software Defined Networks (SDN) and extended Management and Operation (MANO) functions.
As the figure shows, MDs are regular smart phones, wearable gadgets or embedded devices integrated in scooters, motorbikes or cars. They gather sensor data using light IoT protocols like CoAP, and provide access to services of different nature. MIGRATE considers the dynamic creation of virtual functions (VxFs) in the form of vMD to cope with processing and data cache in a MEC fashion, given the energy and computing constraints of MDs. Data collected is then consumed by a data analytics function at a cloud virtualization domain, which feeds final services. Upon the MD movement to a network point of attachment that belongs to a different network virtualization domain, it is necessary to maintain connectivity with its virtual counterpart. MIGRATE bets on the dynamic instantiation of new vMDs on demand, which will be based on the same software platform and will inherit the configuration parameters of the former vMD. To finally make operational the new vMD, data paths will be updated using SDN functions, pointing monitoring and request messages to the new VxF. This is possible since the network entrance point to the virtualization domains are SDN-capable switches. The transfer of data to a newly created vMD is deferred until it is checked that the MD is detected as stable in the new edge virtualization domain.
The associated data base, which is being provided as another VxF, will be maintained synchronized among the different network domains involved. A distributed data base approach has been followed in the solution, using a cluster-based data base management system. Hence, vMD data will be synchronized among data base “copies” through the backbone network, which is expected to provide a high throughput.
Different 5GINFIRE facilities can be involved in the experiment, as the diagram shows. 5G verticals involving mobile devices could benefit from the solution, such as the 5GPP-identified automotive, smart city and healthcare, which in fact has presence within the 5GINFIRE ecosystem. OdinS, as an innovative and technology-based SME with proved experience in these sectors, sees the potential benefits of this solution for providing new mobile services in 5G-powered areas such as transport, in which European 5G corridors are being developed. Hence, a reference implementation of the solution has been developed within the 5GINFIRE ecosystem, including new software modules in the form of VxFs that will remain as facilities to be reused within the project lifetime and beyond. The reference implementation has been evaluated in the IT-Av testbed, but migration is prepared to be performed between sites physically located far away.
With the aim of both validating the MIGRATE platform and assess its performance, the solution was tested under real driving conditions. This was done, as said above, using the vehicular testbed of IT-Av after testing in laboratory the good operation of all modules.
Golf IV 1.9 TDI was used to carry out the tests. This vehicle was driven within the IT-Av premises in Aveiro, taking advantage of 802.11p connectivity provided by the testbed, with two RSUs connected to two different virtualization domains (Open Stack infrastructures). The RSUs has been installed at different sides of the building, and the vehicle moves next to them using the two roads available. Multiple rounds were performed to gather significant results.
The MD mounted in the vehicle has been provided with the monitoring software to access the OBD-II interface and report the data collected. It is in charge of registering with its virtual counterpart (vMD) and then continuously send data to it. This data flow is first locally processed and stored in the Cassandra DB, which automatically is in charge of synchronizing with the rest of DB clusters, including the one setup in 5TONIC. The data analytics module obtains vehicle data from this DB node and forwards it to the Grafana system, with no particular processing algorithm in this case. Hence, end users can finally check past and current status of the vehicle by accessing the Web interface of Grafana.
Given the two 11p RSUs available, when changing the network attachment point, the vMD used is migrated between the two available virtualization domains. The two 11p RSUs are connected with the same router, being part of the same network, and no association is necessary with 11p, so the tests have been done without considering layer two and layer three mobility. Hence, the performance study is focused exclusively on checking the migration approach.
The good operation of the solution was checked initially with the Grafana web interface. The figure bellow shows the results obtained in one of the test rounds performed. The plot panels are configurable, but in this view the vehicle engine load (upper) and coolant temperature (lower) are showed. It can be seen that the test was performed at very low speed according to the engine load. It is worth to mention that the test duration is about two minutes, enough to test the handover process while the data were collected.
The results obtained in one of the testing rounds are plotted in the next figure to showcase the operation of the migration approach. The plots are directly generated from the network analyser Wireshark, using the pcap trace file saved in both MD and vMDs. vMD1 belongs to the source virtualization domain, while vMD2 belongs to the destination domain. As can be seen in the results, MD sends continuous monitoring data to the initial vMD1, considering a previous registration with MD manager. The vehicle maintains more time within the 11p RSU1 coverage due to initially it is stopped. Then it moves and reaches a point where the messages start to be received by 11p RSU2, so a migration event is generated and the data flow is moved to vMD2. As can be seen in the plot, no data losses are recorded. In fact, in all the tests performed, no data losses have been perceived due to the migration solution.
A MANO and SDN-based solution is proposed to implement a state transfer approach for digital twins. In this way, mobile devices are able to maintain MEC capabilities on the move. Upon entering a new virtualization (MEC) domain, a transparent mechanism is used to maintain communication with a former virtual mobile device until a new one is prepared under the new domain. Data synchronization is assured by using a cluster-based database.
The solution has been implemented and deployed in terms of both software and hardware, within the 5GINFRE IT-Av and 5TONIC. A real vehicle has been equipped and driven to validate the solution, at the same time the operation of the migration approach is studied in detail. The results show that a seamless migration between virtualization (MEC) domains is achieved, using a solution that is transparent for both mobile devices and services.