Elastic Content Distribution Based on Unikernels and Change-Point Analysis

Abstract:
The emerging 5G networks call for new approaches to CDNs through addressing challenging issues, such as: (i) scalable and holistic resource management, spanning from large data centers to the user device, including edge clouds; (ii) incorporation of heterogeneous physical and virtual resources; (iii) extensibility to support new capabilities or mechanisms; and (iv) adaptability to dynamic user requirements, server resources and network capacity constraints. User-generated content is the driving force for new services, while edge cloud solutions are being proposed to host (or cache) content locally. However, it is costly to deploy traditional clouds near end-users and such virtual machines (VMs) are inefficient for dynamic network conditions (i.e., may boot-up in minutes). Here, we propose an elastic content distribution platform that serves the Internet content using tiny Unikernel-based VMs. Such VMs are hosting one or a few videos each, appear rapidly in nearby cloud deployments, serve users and then disappear. In other words, the studied environment provides content dissemination through very dynamic, almost ”fluid” VM placement, since the content is packaged with the server software with just a minor increase in size. So, we reposition the content caching and provisioning as a VM orchestration problem. We demonstrate the complete implementation of the proposed platform with proofof- concept experimental results.