Roaming Optimization using Global Presence of AWS: SK Telecom Roaming Edge Cloud

Communication Service Providers (CSPs) are exploring public cloud hosting of network functions, driven by industry transformation. One promising use case is hosting a network function on AWS for outbound roaming subscribers, leveraging AWS’s global infrastructure and enabling enhanced user experience. This post details a reference architecture, including IP Exchange (IPX) connectivity options, allowing integration with existing roaming networks without changes. It also presents test results from SK Telecom’s proof-of-concept deployment of User Plane Function (UPF) and Packet Data Network Gateway User Plane (PGW-U) on AWS global Regions, demonstrating the feasibility and benefits of this approach.

The telecommunications industry is undergoing a transformative shift as CSPs seek to leverage the power of the AWS Cloud. This shift is driven by the desire for cost efficiency, the need for a fully managed service experience, and the exploration of future directions in the industry. In other contexts of AWS, the core network disaster recovery use case and the 5G Stand-Alone (SA) network buildout cases are introduced. As another use case of leveraging the AWS Cloud for telco core networks, a compelling idea is to host core network functions on AWS global Regions to serve international roaming services of CSPs.

International roaming service plays a crucial role in today’s interconnected world by facilitating seamless communication for travelers. As individuals cross international borders, this service ensures uninterrupted connectivity, fostering business collaborations, maintaining personal connections, and providing essential support in emergencies. The significance of this lies in its ability to bridge distances, creating a more accessible and interconnected global community. Despite the fact that most CSPs already provide international roaming service for their subscribers, an approach in this blog proposes a new architecture for the international outbound roaming service for CSPs. This approach harnesses the global infrastructure of AWS, enabling CSPs to provide an improved user experience for their subscribers while they roam internationally. The proposed solution involves deploying network functions, such as the PGW-U and UPF, on AWS global Regions. By utilizing AWS’ distributed infrastructure, CSPs can offer seamless connectivity and services to their roaming subscribers, regardless of their location. This approach not only improves the user experience but also offers scalability and flexibility, allowing CSPs to adapt to changing demands and technological advancements.

In this blog, the details of the reference architecture and live test results conducted by one of the most innovative CSP of the world, SK Telecom (SKT), are demonstrated. This includes options and considerations for establishing connectivity between AWS Virtual Private Cloud (VPC) and IPX providers. This connectivity is crucial for roaming support and forms an essential part of the core network reference architecture on the AWS Cloud. The rest of the blog is structured as follows: first, a quick overview of the 3rd Generation Partnership Project (3GPP) roaming architecture will be presented, followed by an exploration of the reference architecture and technical considerations for VPC, and concluding with a discussion on IPX connectivity. After this, test results from SKT measured in two geographical locations will be demonstrated as proof of concept for the proposed roaming architecture.

Prerequisites – general overview for 3GPP roaming

The 3GPP defines the roaming architecture in the standards with two different scenarios: local breakout (LBO) roaming and home-routed roaming. Figure 1 and 2 show the architecture of home-routed roaming for the 4G and 5G cases, respectively, as defined by referred standards (23.401 for 4G and 23.501 for 5G). In this diagram, HPLMN refers to Home Public Land Mobile Network (PLMN), while VPLMN stands for Visited PLMN, as defined in the standard. LBO roaming refers to the case to break out user traffic at roaming country using a roaming partner’s core network function such as PGW or UPF. Meanwhile, home-route roaming refers to send user traffic back to home country via IPX network, and then break-out traffic to the Internet using home PGW or UPF. Each scenario has its own advantage. More specifically, LBO roaming introduces lower latency driven better user experience of roaming. But it is hard to apply Home PLMN CSP’s own policy since it is using VPLMN partner CSP’s core network functions such as PGW or UPF. On the other hand, home-routed roaming has an advantage to apply HPLMN CSP’s own policy and rule for users (such as rate-control after certain usage amount), but inherently introduces larger delay than the LBO because user traffic has to land at home country of CSP.

In the modern era of global connectivity, international roaming services have become a fundamental offering for CSPs. However, these roaming services also present an opportunity for CSPs to differentiate themselves from competitors. As a result, the practice of home-routed roaming, where outbound roamers’ traffic is routed back to the home network and subjected to the CSP’s policy logic, is more commonly adopted over local breakout methods. By leveraging home-routed roaming, CSPs can apply their own unique policies and optimizations to the roaming traffic, thereby enhancing the user experience and potentially offering differentiated services to their subscribers while they are traveling internationally. This approach allows CSPs to maintain control over the roaming experience and tailor it to their specific requirements, rather than relying solely on the roaming partner’s network capabilities, even though it introduces inevitable latency of service to redirect traffic back to roamer’s home country.

Figure 1. 4G Home-routed Roaming Architecture

Figure 2. 5G Home-routed Roaming Architecture

New architecture proposed by SKT : roaming edge on the AWS Cloud

Reference architecture
When a CSP employs home-routed roaming, user traffic must traverse the overseas link through the IPX network to reach the subscriber’s home country. This can introduce significant propagation delay, depending on the distance between the visited and home countries. To mitigate this issue, SKT designed a new roaming architecture leveraging AWS’ global presence, named Roaming Edge on the AWS Cloud. This approach reroutes user traffic to the AWS Region closest to the visited operator, avoiding inter-countries networks and reducing propagation delay. AWS currently operates in a total of 33 geographic regions worldwide and each AWS Region has multiple, isolated locations known as Availability Zones.

The proposed roaming architecture comprises two distinct patterns. In the first pattern, illustrated in Figure 3, the control plane Network Functions (NFs) are situated within the CSP data center in the home country, while the Packet Data Network Gateway User Plane (PGW-U) and/or User Plane Function (UPF) is deployed across one or more AWS Regions in the visited country. And the VPC that hosts PGW-U and/or UPF is connected to IPX network through AWS Direct Connect (DX). When a subscriber roams to a visited country, all signaling traffic is routed to the home country’s data center via the IPX network. Conversely, user traffic is locally broken out at the nearest AWS Region Internet Gateway (IGW), circumventing the need to traverse overseas links. This approach optimizes the route path, reduces latency, and effectively achieves the same benefits as local breakout roaming in the 3GPP standard while adhering to the CSP’s policy rules.

In the second pattern, depicted in Figure 4, the CSP can leverage the home country’s AWS Region to create a dedicated Packet Data Network Gateway Control Plane (PGW-C) in 4G or Session Management Function (SMF) in 5G for outbound roaming purposes. In both patterns, the PGW-C or SMF is responsible for selecting the proximate PGW-U or UPF to the outbound roamer if multiple PGW-Us/UPFs are deployed across multiple AWS Regions. This geo-proximate PGW-U/UPF selection can be facilitated through the Serving-PLMN (SPLMN) information and User Location Information (ULI) in the message originating from the VPLMN Core Network. The former pattern offers the advantage of leveraging existing PGW-Cs/SMFs in the CSP’s data centers, while the latter pattern allows for the configuration of PGW-U/UPF selection rules at a dedicated PGW-C/SMF in the AWS Cloud.

Figure 3. User Plane only on AWS for optimized-roaming 

Figure 4. Control Plane and User Plane on AWS to optimized-roaming

VPC design
The VPC architecture for the roaming core network on AWS follows the same principles guided in the previous white paper, titled “5G Network Evolution with AWS”.

Figure 5. Roaming Edge VPC reference architecture

Figure 5 illustrates the Roaming Edge VPC reference architecture. For the Proof Of Concept (POC) testing, SKT utilized Samsungs’ cloud-native 4G/5G common core network of PGW-U and UPF. Samsung’s NFs are compatible with running on Amazon Elastic Kubernetes Services (EKS), leveraging Multus support for separating networks per each purpose, such as signaling and user traffic. To ensure high availability of network functions, EKS worker node groups need to be deployed across multiple availability zones.

When hosting PGW-U or UPF on the AWS Region, various Internet breakout models can be considered, as illustrated in the blog for data network breakout on AWS. Since this Roaming Edge on the AWS Cloud aims to provide the shortest latency for the Internet breakout at the local country, the use of an Internet Gateway (IGW) based breakout is required. To enable address translation between User Equipment IP (UE IP) and the public IP address (AWS Elastic IP), AWS NAT Gateway (NAT-GW) can be leveraged. Additionally, the use of NAT-GW disallows direct access to resources in private subnets from the Internet, enhancing the security posture of the solution. As another security measure, when PGW-U/UPF is hosted on AWS and makes Internet breakout through IGW, AWS Shield provides baseline protection against distributed denial of service (DDoS) attacks from the Internet.

Towards the IPX connectivity side (3GPP S8U and N9 interfaces) of PGW-U/UPF, DX via VPN Gateway (VGW) or AWS Transit Gateway (TGW) can be leveraged, as described in the hybrid connectivity whitepaper. Further details on IPX connectivity options and considerations will be provided in the next section.

IPX connectivity using Direct Connect
Establishing interconnectivity between the HPLMN VPC hosting network functions and the IPX provider network is crucial for accommodating NFs for the roaming network. IPX, as defined by the Global System for Mobile Communications (GSMA), is a global, private, multi-service, and secure IP backbone network that connects different service providers. In today’s general 4G/5G roaming (excluding roaming for the Internet of Things), it is necessary to support various multimedia services and ensure reliable connectivity for this interconnection. AWS Direct Connect (DX) can be used to establish a dedicated network connection between an on-premises network and AWS. With a DX connection in place, you can create virtual interfaces directly to the AWS Cloud, bypassing Internet Service Providers (ISPs) networks and ensuring a consistent network experience. However, interconnectivity to the IPX network has a specific requirement to use uniquely routable public IP addresses and public Autonomous System Numbers (ASNs) registered in IR.21. Figures 6 and 7 illustrate the best practice design patterns for setting up DX between the HPLMN VPC and DX to meet this requirement.

Figure 6. DX configuration to IPX – pattern1 using ASN overriding and IR.21 registered IP as VPC IP

Figure 7. DX configuration to IPX – pattern2 using eBGP Multi Hop configuration

This blog discusses two patterns for establishing a DX connection between the IPX network and the Amazon Virtual Private Cloud (VPC) hosting the NF for roaming edge. The first pattern, illustrated in Figure 6, involves setting up Border Gateway Protocol (BGP) peering between the IPX Provider-Edge (PE) router and the AWS Transit Gateway (TGW) using a private Autonomous System Number (ASN). This private ASN is then overridden by the CSP’s ASN registered in IR.21. For the IP address of the NF within the VPC, the CSP’s IR.21 IP address can be used from a private subnet of the VPC, either from the primary VPC CIDR or a secondary CIDR.

As shown in Figure 7, the second pattern involves using eBGP multi-hop (eBGP-MH) to establish a direct BGP connection between the IPX network and the NF within the HPLMN VPC. In this case, the NF inside the HPLMN VPC would use the CSP’s ASN and IR.21 IP address to populate its roaming service IP address directly to the IPX network.

In addition to considering the IR.21 registered IP address and ASN for the DX connection with the IPX network, it is crucial to consider the number of route entries exchanged between the IPX network and AWS DX. Typically, the IPX network advertises all global operators’ roaming networks, which can sometimes be advertised with small CIDR blocks such as /28 or /30, resulting in more than 1,000 route entries being advertised to AWS DX. However, from the perspective of the Roaming Edge VPC in the AWS Region, the path to the IPX network for 3GPP S8 or N9 interfaces can be treated as a default route. Therefore, the IPX network can summarize all roaming networks as a single default route via DX.

Lastly, while two patterns of IPX connectivity are introduced in this blog, there are other options available, such as building an overlay network using a virtual router appliance to logically terminate the IPX network inside AWS VPC rather than at the physical data center of the IPX provider.

Proof of concept test result
SKT has been at the forefront of the mobile industry’s growth since 1984 and has played a crucial role as a global leader by spearheading innovation in communication, cloud, and AI technologies. Throughout 2023, SKT conducted POC testing in two international roaming countries, Spain and Philippines. In this POC, Samsung’s PGW-U and UPF (4G/5G common core) is deployed in the AWS Frankfurt and Hong Kong Regions, with a connection to the BICS‘ IPX network. (Note that, this solution is awarded by WCA 2023 by the name of Roaming Edge Cloud (REC)).

Figure 8. Overview of SKT REC PoC

Using this setup, SKT conducted measurements to evaluate the performance of outbound roaming in terms of measured bandwidth and latency using the Ookla speed test. Furthermore, SKT evaluated real user experience by monitoring global website access speeds (e.g., Netflix and Google) and response time of Google Maps.

As shown in Figure 9, the speed test results confirm improved performance of outbound roaming in terms of throughput, Round Trip Time (RTT), and jitter when the outbound roamer is utilizing the NF hosted on the AWS Region of the visited country. RTT decreased by approximately 84% (from 555ms to 90ms), while the download speed improved by around 30% (from 337 Mbps to 439 Mbps). In addition, as shown in Figure 10, the overall user experience of outbound roaming has demonstrated a more significant and meaningful improvement. For example, the access speed to global websites is reduced by up to 76.9% (Google: from 2.3 seconds to 0.9 seconds, Apple: from 5.7 seconds to 1.3 seconds, Facebook: from 4.0 seconds to 1.4 seconds). In addition, the download time for South Korea’s popular domestic websites integrated with a global content delivery network (CDN) service has been reduced by up to 61.4% (Naver: from 3.2 seconds to 1.9 seconds, Daum: from 13 seconds to 5 seconds), while the response time for searching famous tourist destinations on Google Maps has been improved by up to 43% (from 1.5 seconds to 0.85 seconds).

Figure 9. Speed Test site measurement result at Spain.

Figure 10. User experience improvement by Roaming Edge Cloud solution at Spain. 

Figures 11and 12 show the test results for the Philippines. While the improvement rate is lower compared to Spain, given the close physical distance to South Korea, meaningful enhancements were observed.

Figure 11. Speed Test site measurement result at Philippines.

Figure 12. User experience improvement by Roaming Edge Cloud solution at Philippines.

Conclusions

The live tests conducted by SKT, a leading CSP, have proven the viability of the proposed solution. The test results, measured in different geographical locations, showcased the seamless connectivity and improved performance offered by the Roaming Edge implementation on AWS. By adopting this innovative approach, CSPs can future-proof their networks, enabling them to stay ahead of the curve and meet the evolving demands of the global telecommunications landscape. The successful collaboration between AWS and industry leaders like SKT, Samsung, and BICS paves the way for further advancements and the exploration of new possibilities in the realm of cloud-based core network deployments. For more information about telco 5G use cases on AWS, contact aws.amazon.com/telecom/contact-us.