By Rohini Vugumudi, Hany Fame, Pardeep Singh & Zhen Lu, Comcast
The modulation profiles for Cable Modem Termination Systems (CMTSs) have been historically applied manually, only changing with response to stimuli such as frequency impairments identified by field engineers or, worse, customers. This manual feedback loop is inherently slow, resulting in profile configurations that are limited by the impairments of the lowest common denominator on a given cluster of customers. With the advent of Data Over Cable Service Interface Specification (DOCSIS) 3.1 came the amazing and powerful ability to automatically adapt downstream profiles in near real-time leveraging machine learning and the Profile Management Application (PMA) concept. Tightening and automating the feedback loop allows for the recovery of previously wasted capacity, thereby making the entire network more efficient.
The authors of this publication have developed an implementation of the PMA concept that allows Comcast to manage the downstream and upstream environments efficiently at scale. Our current nascent architecture allows us to run a complete feedback loop every 6 hours; this runtime should only get better with further optimization of the Analytic Engine (AE) service, which is currently the bottleneck. Comcast can now optimize for the best performance, reliability and throughput in an automated and scalable fashion.
In this paper, the service architecture platform for dynamic data collection is called Genome, and it offers a key component of the overall configuration management system. Genome is responsible for the aggregation of data collected from cable modems and other customer devices, and the configuration management service is responsible for the application and validation of generated modulation profiles to their respective parent CMTSs. In particular, the details of adapting modern cloud computing tools to architect a reliable software solution for both downstream and upstream configurations are discussed. There are many details involved in the data aggregation, application of configurations, and validation of configurations, all of which are discussed.