Page 57 - Kaleidoscope Academic Conference Proceedings 2022

TOWARDS COMPUTING AND NETWORK CONVERGENCE:
QOE-ORIENTED SERVICE ANYCAST BASED ON SRV6

Zicheng Wang; Yetong Wang; Xian Gao; Shuai Wu; Wei Lin

Inspur Communications Technology Co., Ltd., China

ABSTRACT tasks can be forwarded to the central cloud for completion,
i.e., a collaborative cloud-edge-terminal system [3].
The new mobile Internet services represented by extended
Reality (XR) have brought new challenges to computing and As Network Functions Virtualization (NFV) is adopted by
networks. A new generation of Information and more and more equipment vendors, various network
Communication Technology (ICT) infrastructure with both equipment including base stations will also have the ability
computing and communication capabilities is being to provide edge computing services. In addition, due to the
promoted. Traditional load balancing technologies cannot transition to cloud-native for all types of application servers,
sense the connection between users and services. It is also a complete application will likely be divided into multiple
difficult to support large-scale distributed computing service decoupled micro-services, which implement business logic
instances, so the service experience provided can be poor. through mutual calls between Application Programming
Interfaces (APIs) [4], and XR-type applications are no
In this paper a service anycast system based on a Segment exception. These micro-services will be distributed on
Routing over IPv6 (SRv6) data plane is proposed, which can different computing nodes, including edge computing nodes
provide a Quality of Experience (QoE)-oriented load and central cloud computing nodes, etc. In the foreseeable
balancing capability for the network according to the future, computing infrastructures will be distributed in large
obtained status information of network and computing numbers over a huge spatial scale [5], XR-type application
service instances. Besides, an IP address identification service instances can be deployed at all or part of the micro-
mechanism is also proposed to help the control plane handle service components in a distributed manner, and user
traffic engineering policies, and to support efficient service terminals often access XR-type applications on the move.
resource discovery, seamless mobility, and service This leads to a high degree of time variability in the service
continuity. instance status and the link between the user and the service
instance. The previous DNS-based load balancing
Keywords – Anycast, Computing and Network mechanisms [6] are difficult to fully utilize the decentralized
Convergence (CNC), load balance, SRv6 deployment of computing resources to provide good QoE to
users because they do not sense the network status between
1. INTRODUCTION users and service instances. However, the load balancing
solution based on the service status list maintained by the
With the global promotion and deployment of 5G networks, client requires terminal devices to collect service node status
the large bandwidth and low latency wireless connectivity information and measure network status information in real
provided has brought new opportunities for the development time. This introduces a large resource overhead for low-
of interactive multimedia applications [1]. New immersive power devices, such as Internet of Things (IoT) devices.
and interactive applications represented by XR not only need Since the cloud-edge-terminal collaboration process of
to collect and process feedback data from various sensors in various applications, including XR, will be abstracted from
real time to evaluate the information including user's position, the network's perspective as the mutual invocation of
speed, perspective, but also need to quickly render highly multiple micro-service components between user terminals
realistic images with high resolution based on this and micro-service clusters and across micro-service clusters,
information [2]. Therefore, the rapid rise of XR-type which provides a potential solution for processing node and
applications not only puts the network's service assurance link selection among a large number of service nodes and
capability to the test while also posing new challenges to the guarantees service experience quality, that is, load balancing
computing performance of terminals that are highly through the control plane of the network.
constrained in terms of power consumption and volume. It is
a feasible solution to use edge computing to offload real-time In this paper an overlay service anycast system based on the
workloads on terminals with extremely low service SRv6 protocol is proposed, which utilizes the underlay
processing latency, while some non-real-time computing routing and traffic engineering functions provided by the

978-92-61-36151-8/CFP2268P @ ITU 2022 – 11 – Kaleidoscope