Data Efficient NSX Inventory State Management in Cross-Cloud Environment

Abstract

In a typical NSX deployment, software assets such as virtual machines and virtual network interfaces of a distributed set of ESXi / KVM hosts need to be discovered, collated and persisted. In traditional on premise deployments, as the scale of these host elements grows, it is impractical to persist all the attributes of the interested asset set on the NSX manager. It is more efficient to hold some of the attributes in memory and fetch them again in case the NSX manager crashes and recovers. The NSX manager itself is deployed in a high availability configuration with shared storage, thereby reducing the risk of loss of data. This product plans to support scenarios where these assumptions of repopulating the entire data set may not be cost effective in future. Take for instance a cross cloud / hybrid cloud deployment scenario, where the NSX Manager is deployed on the premises of an enterprise customer, which is managing SDN use cases for both on premise and public cloud workloads. In such a situation, inventory data must be collected from the public cloud environment and sent over WAN links to the on premise NSX manager. The product must evolve, to minimize the need for repopulating host inventory information as it encounters various faults. The aim is to study the full sync scenarios and find the current size of inventory information for a typical NSX deployment and propose a solution to effectively reduce the inventory delta exchange.