A Scalable Heterogeneous Software Architecture for the Telecommunication Cloud

摘要

In the recent years telecommunication operators have been more and more demanding for a cloud-native deployment on their already existing Information Technology (IT) infrastructure of the IP Multimedia Subsystem (IMS) core and the telecommunication platform in general. Such flexibility requires the telecommunication vendors to develop new software architectures allowing the deployment of many different cloud platforms or even heterogeneous cloud platforms automatically in a unified fashion.;This thesis proposes a cloud-native, scalable, heterogeneous and distributed software architecture based on the actor model. The architecture allows development of software once and a single software version on multiple cloud platforms, e.g., bare metal Linux servers, IaaS, PaaS, etc. It enables solution-based and hybrid cloud deployments. A novel concept of pouch is proposed to abstract the deployment platform from the application. The architecture is based on the adaptation of a number of cloud architectural patterns. The horizontal scaling and sharding patterns are used to distribute the processing on pouches while maximizing cache hit. The auto-scaling and busy signal patterns ensure QoS by scaling out when capacity of a pouch is reached. MapReduce pattern facilitate collection of logging information. The collocate pattern is applied to minimize the latency experienced by the application. Finally, usage of the actor model makes the proposed architecture highly granular, allowing a persubscriber, per-service deployment.;The proposed architecture allows efficient resource allocation through the cloud in order to prevent overloading of computing resources and the resulting adverse impact on QoS. It also allows application state information to be distributed on compute instances, providing resiliency while minimizing impact on latency. The proposed architecture was implemented as a framework which can be reused for any distributed cloud application. We built on that framework a simplified IMS Core system to demonstrate how IMS applications can be re-engineered to be provided as cloud-native applications. Experimental results show the architecture to be viable and having comparable characteristics, linearly increasing with the number of users, on different cloud platforms. Moreover the results obtained on a bare-metal platform, and in an OpenStack-based virtual environment show that a cloud-based, distributed approach has benefits over the traditional node-based architecture in terms of automated scalability, minimizing human intervention need and potential errors in dimensioning and in terms of elasticity.