Service Routing in Multi-access Edge Computing

The operators are deploying Multi-access Edge Computing (MEC) to provide services with lower latency to their users. Comparing to accessing service in the clouds, the MECs can provide service much nearer to the users. However, in the current architecture of Internet, we need to send a DNS query to get the IP address of the service firstly, and then access the service . It is not the optimal solution in the MEC scenarios which are sensitive to the latency of service accessing. In this document, we introduce a mechanism that can access the service directly without the DNS procedure. In the 5G architecture, a UE (User Equipment) needs to connect to a UPF (User Plane Function) working as a gateway by using a tunnel, and then access service via the destination IP address. In the scenarios of MEC, the service may be accessed within the MEC, meanwhile the MEC also provides a UPF Function for the UEs. Therefore, in fact, the service access takes place in a limited domain . In this limited domain, we can use a specific IP address to directly access the service.

In the proposed mechanism, a UE should have a session with the UPF in the MEC. Also, the UE should be aware that it can access the service more quickly within the MEC if the service is available in the MEC. The proposed mechanism is described briefly as below. Firstly, the UE sends a normal DNS query to the attached MEC if it wants to access a service, such as "www.local-weather.com". Meanwhile, it can send a connection establishment request for the service to the attached MEC, and try to establish a TCP/QUIC connection directly. In the request, the destination IP address is a specific IP made by the UE itself by hashing the domain name. Secondly, the UE may establish the connection successfully by using the specific IP address, and get access to the service bypassing the DNS procedure. It will take place when the UE receives the response of the connection establishment request before receiving the response of the normal DNS query. If the DNS response returns firstly, the UE will do the normal service access procedure. It means that if the UE fails to establish a connection using the specific IP firstly, the UE can wait for the normal destination IP address received from the DNS procedure. In this mechanism, the specific IP address can contain some information about the service, so we call it service routing in this document. The specific IP address is called the Service Routing IP address.

There are several options for the Service Routing IP address. The address has the same structure as the IPv6 address defined in . In the first option, we can assume that the UE can receive an MEC prefix for the service routing in the procedure of establishing the session between the UE and the UPF in the MEC. For example, the length of an MEC prefix is 64 bits, and the length of the hashed domain name is also 64 bits. In the MEC, the server of the service should use the same hash algorithm to generate the Service Routing IP address, and the 128 bits IPv6 address should be routed correctly within the MEC. Hence, the MEC works like a virtual network node containing services, with the MEC prefix as a Location, and the hashed domain name as a Function. In the second option, we can use a ULA IP address (Unique Local Address) for the Service Routing IP address . The procedure is similar to the first option, but the Service Routing IP address becomes the format of <MEC_ULA_Preifx: Hashed_Domain_Name>. The MEC_ULA_Prefix contains a specific subnet-ID. In the last option, we can use all the 128 bits as the Hashed_Domain_Name. In this situation, the UE does not need to receive a specific prefix in advanced, and all the services in different MECs have the same IP address for the same service to support this quick access.

In the MEC, the network should support forwarding the Service Routing IP. In the client and server, they should support the binding of the Service Routing IP and the traditional DA IP. The value of the Service Routing IP exists mainly in the period of establishing the connection. After the connection is established, we can use the normal DA IP instead. In the mechanism of this document, the MEC will receive a normal DNS query, and a connection establishment request for the service based on service routing. The MEC will try to establish the connection directly with the UE. Meanwhile, the MEC also does the normal DNS procedure for the UE. They take place independently, so that after the procedure of DNS, the MEC will response a target IP address to the UE no matter whether the connection establishment successes or fails by using the Service Routing IP address.

In QUIC , there is a "Server's Preferred Address" mechanism. Perhaps it can help the DA changing process. QUIC allows servers to accept connections on one IP address and attempt to transfer these connections to a more preferred address shortly after the handshake. We assume that the mechanism about the "Server's Preferred Address" is supported both in the client and server, and the connection is a QUIC connection. Thus, the UE can use the hashed DA address to establish the connection, and after that, use the Server's Preferred Address instead. In this situation, the Server's Preferred Address should be the same as the normal DA IP address obtained in the DNS process mentioned before.

At the beginning of the adoption of the mechanism, we do not think there would be too many essential services requiring this ultimate user experience, so that we assume that there would be no Hash conflict between the services. Besides, if there is any conflict in the MEC, the MEC can find it before deploying the service. If the mechanism is adopted widely, and conflict exists between hashed domain names in the MEC, we can enable the mechanism only on the most essential service. Another option is to change the HASH algorithm that is running on the clients and severs to make a better Hash result.

MEC can also support accessing via fixed clients. In this situation, the BNG (Broadband Network Gateway) as the gateway of the client can work similarly to the UPF. A tunnel between the BNG and the MEC may be needed, and the MEC prefix can be obtained in the procedure of authentication. In the authentication of a fixed client, a more static session can be established because the client will not move.

TBD.