LISP-Based Control Plane for Service Connectivity in Multi-Cluster Cloud Systems

Kyoungjae Sun,Younghan Kim,Jangwon Lee

doi:10.1109/access.2022.3155113

Kyoungjae Sun, Younghan Kim + Show 1 more

Open Access

https://doi.org/10.1109/access.2022.3155113

Copy DOI

Abstract

In mobile networks, the edge cloud environment has emerged to provide various services to users. When the mobile user moves to another location while connecting the edge cloud service, to maintain an optimal service path, it is necessary to migrate the service to a new edge cloud close to the location where the user is newly connected. Several methods have been proposed to integrate edge cloud service migration technology with existing mobility management protocols in mobile networks, and a location/ID separation protocol (LISP)-based service optimal path management method between edge clouds was recently proposed. However, for real implementation of edge cloud with existing platform specified to containerized infrastructure, where the address of service is allocated randomly and locally, it is hard to discover appropriate service from the external client as well as maintain service connectivity when service location is changed. To solve that, this paper proposed the LISP-based service ID management system which provides uniform access for equivalent services running on different edge clouds. To apply the proposed ID management system on real distributed edge cloud systems, the mobility management and edge cloud networking systems were integrated as shown in the testbed implementation. This shows the possibility of combining the isolated mobility management for mobile users and internal identification of edge services using LISP for reducing the interruption and delay of edge service for mobile users. As a result of the experiments conducted on a real testbed, our proposed system was verified to enable a change in the routing path while maintaining a single service ID between different edge clouds, and the delay time for path reconfiguration is reduced compared to the existing method.

Highlights

Distributed edge cloud environments have the advantage of minimizing the delay in accessing services and distributing traffic by deploying virtualized services close to user access locations [1]
By managing an IP address pool for the service objects initialized by each cluster, the location/ID separation protocol (LISP)-based control plane provides uniform service IDs to external clients, regardless of the location of a service
When a packet destined for a service ID address is sent from an external client, it can be delivered to the correct pod through tunneling between LISP xTRs, which is transparent to the current Kubernetes networking model using the container network interface (CNI) plugin

Summary

INTRODUCTION

Distributed edge cloud environments have the advantage of minimizing the delay in accessing services and distributing traffic by deploying virtualized services close to user access locations [1]. According to the basic design principles of Kubernetes networking, to improve the flexibility, the IP address assigned to a pod is non-permanent, and the IP address is changed when a pod is reinitialized [14] For this reason, a fixed endpoint cannot be supported when the service is dynamically instantiated or moved to another edge cloud. A common method used in current systems for exposing services to external clients is to expose services using the public IP address of the gateway or load balancer of the cluster, and translate those addresses into the internal IP addresses of the services [16, 17] In such an environment, a user can acquire the proper IP address for a target service by using a DNS server that records the service names and corresponding IP addresses.

Related Work

IMPLEMENTATION AND ANALYSIS

TESTBED EVALUATION

Findings

CONCLUSION