Implementing Next-gen NF Placement Strategies Today
For CSPs deploying 5G who now realize that they need to lay the foundation for dynamic NF placement system today, as they start rolling out 5G services, where and how should they start?
CSPs will first want to examine the monetization model. CSPs need to model how network slices will be monetized. Further, CSPs will want to locate the integration points between the billing systems, and charging systems and the orchestration systems while ensuring that the appropriate APIs and integration points are tied together.
Next, AvidThink recommends rolling out in three major phases: static, adaptive, and autonomous.
Phase One — Static Placement
Initially, the CSP should pick popular use cases to lay the groundwork. For instance, a disaggregated and virtualized RAN could be an initial use case. This allows them to ease into laying a foundation for our new orchestration system.
As a start, the CSP could enable eMBB with service guarantees around the slice. Static decisions could include where to place the central unit (CU) and distributed units (DUs) in RAN rollout. As well, arrangements will be needed as to the placement of the user-plane functions (UPFs) that will be processing the packets. For instance, UPFs will generally be distributed as close to the UEs as possible to reduce latency. Inventory systems that can capture the attributes of each network component will be necessary to design end-to-end slices that provide the appropriate SLAs.
During this phase, we will incorporate the telco edge, which are ideal locations for CU/DU placements. Using a concrete use case with virtualized RAN allows the carriers to build a sense of which non-telco clouds and locations could be a fit and how to characterize and catalog these non-owned locations. This is also an opportunity to understand the integration APIs present in each location and how best to query available resources and sites in public clouds. It also allows CSPs to examine the feasibility of having EPA-type visibility into the underlying infrastructure capabilities of a public cloud and public edge service. This is when CSPs would assess whether NFs can be appropriately placed onto resources supporting their unique I/O requirements and whether the performance characterization of these public cloud resources is sufficiently reliable to meet strict SLAs. Some public clouds do not provide tight SLAs at the network level, making them unusable for most network slicing use cases.
Other learnings and actions in this phase could include 4G/5G integration. Orchestrating and running a dual-core network (4G LTE EPC and 5GC) will certainly involve managing both containers and VMs (and even PNFs) and allow the telco to evaluate the capabilities of the orchestration systems.
Once the underlying RAN systems are up and running, the CSP can introduce a few additional network slices and start incorporating or local-breakout services, to ascertain that the orchestration system is robust. At this point, CSPs will want to test out the orchestration systems' design-time support in planning out slices with the appropriate SLAs (based on measurements of latencies and performance of the underlying infrastructure). Part of the design-time capabilities should include the ability to play out what-if scenarios to get an idea of how different services could be configured along with some idea of the costs associated. This will allow CSPs to balance costs versus reliability or control versus costs (e.g., using owned resources versus maybe cheaper public cloud resources).
Once we’ve ascertained that the orchestration system continues to perform well, we can take on offering real-time services like connected vehicles, fleet and logistics use cases, and eventually autonomous driving.
Subsequently, the CSP will want to validate the system's run-time support to ensure that SLAs are in fact met, and that the system is able to take into account real-time feeds from the inventory system to account for all that is happening in the network with respect to resource availability.
Finally, and in preparation for the next phase, CSPs will want to verify that the charging system integration is working appropriately and that accurate KPIs and SLA measurements are being fed back into the OSS systems. This would also be the right time to perform pre-work to lay the foundation for real-time on-demand charging for dynamic slices in the upcoming phase.
