In a previous blog, we discussed how artificial intelligence and operations analytics (AI/OA) relate to virtualized networks. In this article, we present five use cases directed to cable operators. The first two apply to cloud computing; and the next two, to the access network. The final is an emerging category. They all drive a variety of resource allocation optimizations in networks that are or could be virtualized.
A cloud-based user interface no longer depends upon the compute resources of a set-top box. Yet if overwhelmed by requests, that system will generate denials for access. What then?
The worst approach is to wait until those denials are issued and customer complaints begin hitting the care staff. A less naïve approach involves waiting until the resources hit a certain threshold and then adding CPU capacity. The ideal approach would be right-sizing a dynamic threshold for given portions of the network with the goal of optimizing resource allocation and only adding virtual machine (VM) capacity when truly needed. It would also keep an eye on real subscriber QoE measures.
Figure 1 shows a variety of data which might be applied to determining when to add VMs to a cloud-based guide service. It also illustrates how an OA/AI system can utilize an orchestration layer to spin up/down VM instances as indicated by the feature-driven model.
Operators have deployed cloud or network DVR service to reduce costs and provide greater value to subscribers. Business models may involve a static storage threshold or deletion of content after a certain number of days.
A more dynamic, AI-driven approach will look at recording trends and assess the need for additional capacity on the basis of virtual disk capacity and variables such as file-size distribution, delete history and recording frequency. At the macro level, OA also could guide operators on the allocation of resources within clusters of VMs and determine which services get how many CPUs, and when.
DOCSIS Channel Licenses
With the emergence of the converged cable access platform (CCAP) and massive channel densities, most CCAP vendors have moved to a license-based model for selling DOCSIS channels. They also have allowed operators to pool licenses across their networks. This allows operators flexibility but raises a challenge – where are these licenses best deployed?
This situation presents an opportunity for AI/OA to enable license-based SDN within the access network. An operator can optimize license allocations, such that licenses from under-utilized portions of the network can be reallocated to “hot spots” elsewhere. Whereas a status quo approach might rely upon one-time historic generalizations, OA can identify real-time variations from narrower sub-populations and prescribe precise and ongoing resource reallocation solutions with much greater efficiency.
In DOCSIS 3.1, Orthogonal Frequency Division Multiplexing (OFDM) enables thousands of QAM sub-carriers per channel, each with its own profile, or modulation value and amplitude. However, there are only 16 OFDM profiles per downstream channel and each cable modem (CM) has its own RF characteristics. Optimizing this limited set of profiles for a large number of CMs per channel can be a difficult task – especially on a CCAP which has limited memory and processing power.
An Operations Analytics system running on general-purpose hardware with not only greater compute and storage capabilities, but also a lower hardware cost can provide a better solution. This external OA/MI system can leverage more RF history and data from other sources. For example, leveraging the usage patterns of individual CMs or data on diurnal/seasonal RF variation, etc. By bringing more data and analytics to bear on the problem, the OFDM profiles can be better optimized for the predicted usage and conditions
A new approach involves the extension of SDN-like concepts into the Operations Support Systems (OSS) arena. This begins by rethinking key OSS systems such as the interactive voice response (IVR) system. Just as a router in the data networking world routes packets, an IVR system routes calls. And just as SDN concepts can be applied to a router, they can also be applied to an IVR system. Thus, we can separate the data plane (i.e. the calls) from the control plane (i.e. the routing decisions based on policy, etc) within an IVR just as we can separate the data plane (i.e. packets) from the control plane in a router or switch. We call this broad application of SDN concepts “Software-Defined Operations.”
Guavus is a leader in the use of AI/OA for Software Defined Operations. A major cable operator used Guavus Live Ops to detect network anomalies and automatically install call deflections within their IVR to inform subscribers of a known or likely outage. This led to $6.7 million in savings from deflected calls – eliminating both unnecessary support calls and truck rolls.
More to Come
Network service providers benefit from advanced analytics today. Leveraging AI, they can better allocate network resources and capacity, predict and prevent customer-impacting incidents, and actively work to maintain a consistent customer experience. With ever increasing deployments of SDN and NFV, service providers will find analytics even more critical to operating their businesses efficiently and effectively. Guavus’ advanced analytics maximize SDN/NFV investments, meeting the needs of service providers today and in the future.
Note: This and the related article were drawn from “Leveraging Machine Intelligence and Operations Analytics to Assure Virtualized Networks and Services,” presented at the Fall Technical Forum, Cable-Tec Expo, 2017.