Topology and Dynamic Resource Availability

For optimal NF placements, the orchestration system will need to take network topologies into account. It will need to understand topological distances and capacity between the user equipment and other mobile devices and where the critical network resources live. Real-time feeds that provide ongoing latency measurements will also be a valuable input into the placement decision. Further, the system needs a real-time inventory of the different edge and cloud network domains it can provision NFs at: from core public or telco clouds to regional resources to local edge resources. In our example above, without a real-time inventory, the orchestration system might not know that some of those edge or cloud locations were damaged in the natural disaster and are out of commission.

Understanding the cost numbers and available inventory — CPU, storage, network elements — in real-time across each of these locations, as well as any metadata around data placement restrictions or SLAs promised, will be crucial. As the system becomes more sophisticated, understanding calendared resource reservations in these locations and perhaps building a predictive model through AI/ML may improve the ability to meet SLAs and control costs.

Application Integration and Support

The next set of considerations include understanding and taking into account the composition and placement of NFs necessary to support the functionality and SLAs that applications require. As part of 5G business models, CSPs expect to provide access to slices with distinct SLAs, or perhaps the ability for enterprises and application providers to create virtual network slices via API calls.

Particularly for low-latency applications, or applications for which strict SLAs are needed, CSPs will need to dig deeper into understanding how to best to support such enterprise applications. For instance, in our use case above, the underlying network needs to support the video feed between the ambulance and the ER with high reliability. If the network is still not able to support the SLAs by using up all available capacity and NF-hosting resources, then the network will need methods to signal to applications to take appropriate action.

Likewise, the expected lifetime or duration of the slice, whether scheduled and known in advance or predicted based on history, will determine what resources are selected and used to improve the likelihood of meeting demands from other applications. This could become a parameter in network slice provisioning.

Finally, knowing how impairment of the network and underlying resources affect the user experience can aid in placement decisions. There could be optimizations that allow the relaxation of constraints without hurting the quality of experience — still meeting the stated SLA but not over-delivering unnecessarily.

NF Characteristics

Once the NFs are identified as part of the application, the orchestration system will need a list of resource requirements for each component: base resource consumption, max consumption and multitenancy requirements. Specific underlying infrastructure needs — hardware crypto assist, accelerated software and hardware networking stacks, GPUs, FPGAs — would have to be taken into account. This is where our experience in the first generation of NF placement will be factored in.

Any specific topological needs for connectivity between NF components will also need to be evaluated. Other considerations include startup times and shutdown and clean-up procedures, which can impact the mobility of NFs if they need to be instantiated or re-instantiated on the fly during the life of the network slice.

And finally, any unique characteristics of the NF that support scalability and load-balancing must also be included. The placement function needs to comprehend the type of state that's kept in NF components and how that state may be shared or transferred during an NF migration, or a failover scenario.

User and Device Characteristics: Volume and Mobility

As part of creating or managing a slice and instantiating the appropriate NFs to support that slice, the orchestration system will need to understand the volume and mobility characteristics of the user equipment (UE) involved. 5G is targeted to support high-density mMTC deployments with up to a million devices per square kilometer. Likewise, it's meant to support high-speed mobility up to 500 km per hour (think high-speed trains).

Therefore, information around which UEs will be mobile, the likely area of coverage needed, and the rate at which the UEs will be moving (e.g., walking versus driving) will need to factor into the overall orchestration and placement decisions. NF placement systems will likely take into account the potential density of use or areas of mobility and pre-provision (or at least reserve resources for) appropriate NFs needed to support the different volumes and types of UE traffic. In some cases, these may not be known ahead of time, and predictive analytics, along with UE characteristics/role information will need to be utilized.