Architecting Gen 2 NF Placement Solutions

To build these second-generation NF placement solutions will require a significant step up from the previous generation of NFV orchestrators. CSPs will need to research and make early investments on this front in parallel with any of their 5G efforts. Simultaneously, CSPs have to innovate on their business models for offering differentiated services with 5G. What type of differentiated services are their business customers asking for? How should they utilize network slicing technology, and how will they package these slices for sale to businesses? What type of KPI and SLA monitoring will they need to put in place, and how will that be communicated to the business customer?

There are many questions, which all require answers to ensure that CSPs are able to extract the ROI from their massive 5G infrastructure investment. Simultaneously, networking engineering teams need to build and architect a 5G orchestration system that provides sufficient flexibility to their business counterparts. 5G is new, and AvidThink anticipates CSPs will run through a few iterations of product offerings before clear business model winners emerge. During this period of trial and error, the 5G orchestration system needs to be sufficiently versatile to support different models.

In the meantime, here are what AvidThink believes to be vital elements of this Gen 2 orchestration:

• Business integration context — The orchestration platform needs to provide the necessary context and intelligence needed to better integrate this new generation of dynamic NF capabilities into existing OSS and BSS. As we described above, we will need to effectively translate the business inputs that define a differentiated service into the underlying parameters for the orchestration system.
• An intent-based, declarative framework — The system needs to have sufficient intelligence to "do the right thing" given a set of constraints, resource availability, cost models, and the target SLA. CSPs will not be able to use a rules-based programmatic approach given a large number of parameters and factors. As such, the input to the system will be high-level, and it will be left to work out how to achieve its goals. Given the complexity of the problem space, it is most likely that AI/ML techniques will be utilized.
• Compatibility across multiple clouds — The system must manage across multiple clouds: on-prem telco clouds at both the core and the edge as well as public clouds. And if public clouds are used at the edge, those need to be accommodated too. APIs necessary to integrate into these clouds to query, provision, and monitor resources will need to be supported by the NF orchestration system.
• Support for multiple platforms and VIMs — The orchestration systems need to span PNFs, VNFs, and CNFs. Further, the solution must accommodate various combinations of bare metal platforms, VM-based infrastructure managers, like OpenStack and VMware vCloud, as well as Kubernetes for containers. There will also be proprietary infrastructure managers in public clouds. The orchestration system also needs to have a scheduling and workload placement system that understands the concept of an overall network service that's separate from the underlying VIMs (which do not semantically understand what a network service is).
• Ongoing closed-loop monitoring and dynamic adaptation — As the underlying infrastructure changes (or fails), new NF instances must be procured and instantiated. The state may need to be migrated and data paths recreated. The NF orchestration system needs to monitor and integrate ongoing telemetry and other underlying information from the infrastructure. It needs to dynamically adapt to maintain the promised SLAs. An understanding of how the telemetry information correlates with service availability and the quality of service will need to be built into the orchestration system.
• Rich NF marketplace or catalog — These orchestration systems will have a wide range of infrastructure to run on and likely be faced with different constraints at different times. To the extent that multiple NFs can fulfill the necessary feature requirements but that have different performance characteristics with various licensing and cost models, the orchestration system will have more options. While there could be increased complexity, the improved flexibility could make achieving the SLAs more feasible.