Telecom IQ
cancel
Showing results for 
Search instead for 
Did you mean: 

NFV Nuts and Bolts: Distributed vCPE - deployment and provisioning aspects in context of OpenStack

TiffanyAlvano

By Senthil Kumar Subramaniam

Lead Architect – NFV Solutions

 

Communication Service Providers (CSP) are actively evaluating NFV and actively engaging in ETSI use case(s) trials. Virtual CPE (Virtual Customer Premise Equipment) is one of those use cases.

There are two ways to implement vCPE:

 

- Centralized – In this method the CPE functions are deployed as VNFs (Virtual Network Function) on a compute hardware deployed within the carriers/service provider datacenter.

 

- Distributed/Hybrid – In this method some of the CPE functions are deployed as VNFs on a compute hardware installed at the customer`s premise and some are deployed on carriers/service provider datacenter. vCPE Centralized deployment is relatively easy to manage compared to the distributed model but the later provides more flexibility and higher level of isolation since the hardware is not shared among customers. The reason for the complexity in the distributed model is due the large number compute nodes (CPE), possibly hundreds of thousands, which need to be managed by VIM (Virtualized Infrastructure Manager). Since OpenStack is the de facto standard for the VIM layer, it is important to analyze the challenges in implementing distributed vCPE. There are multiple aspects to this problem, including:

  • Deployment Architecture
  • Provisioning of the CPE node
  • Provisioning of the VNFs
  • Management of the vCPE functions as a service

The deployment architecture should take into account how the CPE nodes and their OpenStack Controllers are deployed. The standard model of single cluster controllers may not be able to manage large number of CPE compute nodes. A different approach would be the requirement for controller of controllers, where the controller functionality is distributed across different NFVI-PoPs (Regional DC) and the CPE nodes are connected to these controllers. A global controller located at Global DC would in turn need to control the regional controllers. The Regional controller cluster can have a minimal set of OpenStack services compared to the Global DC Controller cluster.

 

The next aspect is the Provisioning of the CPE node which deals with the actual installation (truck roll) of the CPE at customer premises. There are two cases here – CPE comes with a pre-installed image of OS and OpenStack components. In the former case, the regional controller cluster should be able to automatically discover the CPE node. In the later case, it is more complicated and requires iPXE type remote boot frameworks to provision the CPE node. iPXE provides additional capabilities over standard PXE by supporting boot over HTTP and boot over WAN. Once the discovery and image provisioning is completed, the CPE node`s meta data needs to be synchronized with upstream systems inventory (NFV-Orchestrator, Global OpenStack Controller) in order to create a mapping between the Customer ID and CPE ID. Either the technician may manually update this data using a BSS portal, or the provisioned image could have inbuilt logic to do this automatically.

 

At this stage, the CPE node is ready for on demand provisioning of CPE functions. The customer can order VNFs like Firewall, WAN Optimizer etc. It is important to ensure the CPE which is modelled as a compute node, is not used by OpenStack Nova scheduler as part of its standard scheduling of compute nodes. There are multiple ways to solve this issue – one such option is to use Blazar (Resource Reservation Framework). When the customer places an order for a VNF, the correct CPE node should be chosen for the VNF provisioning. Operator`s BSS systems could pass the Customer ID and CPE ID as part of the VNF order passed to NFV Orchestrator which in turn could use the Blazar Reservation ID or Custom Nova filters to choose the CPE node.

 

In summary, the stock OpenStack requires several enhancements in order to support a scalable and yet manageable distributed vCPE NFV solution.   For more information on how HP is solving the vCPE deployment challenges please refer to the vCPE White Paper.          

 

           

 

0 Kudos
About the Author

TiffanyAlvano

Tiffany Alvano works for HP in the NFV Business Unit and helps drive the overall marketing strategy. Tiffany is a social media professional and has been blogging since the start of NFV.

Comments

Simple and Good !

Labels
Events
28-30 November
Madrid, Spain
Discover 2017 Madrid
Join us for Hewlett Packard Enterprise Discover 2017 Madrid, taking place 28-30 November at the Feria de Madrid Convention Center
Read more
See posts for dates
Online
HPE Webinars - 2017
Find out about this year's live broadcasts and on-demand webinars.
Read more
View all