scope:

The present document describes a study of how today's Cloud/Data Centre techniques can be adapted to achieve scalability, efficiency, and reliability in NFV environments. These techniques are designed for managing shared processing state with low-latency and high-availability requirements. They are shown to be application-independent that can be applied generally, rather than have each VNF use its own idiosyncratic method for meeting these goals. Although an individual VNF could manage its own scale and replication, the techniques described here require a single coherent manager, such as an orchestrator, to manage the scale and capacity of many disparate VNFs. Today's IT/Cloud Data Centres exhibit very high availability levels by limiting the amount of unique state in a single element and creating a virtual network function from a number of small replicated components whose functional capacity can be scaled in and out by adjusting the running number of components. Reliability and availability for these type of VNFs is provided by a number of small replicated components. When an individual component fails, little state is lost and the overall VNF experiences minimal change in functional capacity. Capacity failures can be recovered by instantiating additional components. The present document considers a variety of use cases, involving differing levels of shared state and different reliability requirements; each case is explored for application-independent ways to manage state, react to failures, and respond to increased load. The intent of the present document is to demonstrate the feasibility of these techniques for achieving high availability for VNFs and provide guidance on Best Practices for scale out system architectures for the management of reliability. As such, the architectures described in the present document are strictly illustrative in nature.

Accordingly, the scope of the present document is stated as follows:

• Provide an overview of how such architectures are currently deployed in Cloud/Data Centres.

• Describe various categories of state and how scaling state can be managed.

• Describe scale-out techniques for instantiating new VNFs in a single location where failures have occurred or unexpected traffic surges have been experienced. Scale-out may be done over multiple servers within a location or in a server in the same rack or cluster within any given location. Scaling out over servers in multiple locations can be investigated in follow-up studies.

• Develop guidelines for monitoring state such that suitable requirements for controlling elements (e.g. orchestrator) can be formalized in follow-up studies.