Many NFV use cases (i.e. vCPE, vBRAS and vBNG), all require dynamic scaling, agile resource balancing, and geo redundancy across multiple PoPs (Point of Presence). We argue that the current OpenStack platform does not handle network dynamics well because its resource allocation is not network-aware but compute-centric. To overcome this, the network fabric needs to be enhanced to be a dynamic platform that can react to failure, resource contention, and performance degradation in the system. In this talk, we present and demonstrate an OpenStack enhanced dynamic network fabric that offers high availability, manageability, and performance required for carrier-grade. To react to network dynamics we provide underlay network aware PoP level rerouting. Within the PoP, we find the best resource reallocation via bandwidth-guaranteed VM placement. In addition, we use SLA verification to check if the reactions satisfy the SLAs.
This presentation will introduce our experience of building a vCPE PoC on top of OpenStack with production grade VNFs. Our key differentiator to existing platform is a dynamic network fabric that can be aware of the underlay to make deployment decisions based on changes happening in the underlay. We will show that an intelligent overlay routing and VM placement mechanism are in place to guarantee high performance and availability when workloads are relocated and the underlay network changes. To bookend these features, a dynamic SLA assurance engine will be shown to catch SLA violations. The presentation will show the design and algorithms for these technologies and show the benefit of significant savings in vCPE resources.