In 2019, Comcast developed a Profile Management Application (PMA) system for generating and transacting D3.1 downstream (DS) profiles tailored to the conditions of each Orthogonal Frequency Division Multiplexed (OFDM) channel in its network. The approach, machine learning algorithms and system architecture were described in a previous SCTE technical paper. The initial plan for this follow-up paper was to focus on Comcast’s PMA deployment journey, the success of which is evidenced by thousands of Cable Modem Termination Systems (CMTSs) managed by the PMA, yielding greater than 20 Tbps of added downstream (DS) capacity to the network.
With the onset of the COVID-19 crisis, some of that focus shifted, in lockstep with the shift of the U.S. and worldwide workforce from office to home. Figure 1 shows the 32% increase in upstream (US) traffic, post-COVID, and the shift in peak times for DS traffic from 9:00 PM to 7:30 PM, and from 9:00 PM to 8:00 AM & 6:00 PM for US traffic. Figure 2 shows the bandwidth demand growth, around time of the COVID crisis (Spring of 2020), for US traffic (black curve) and DS traffic (sky blue curve). With work from-home traffic increasing on the network, and because of the earlier implementation of the PMA, the DS capacity was available, and the network was able to easily scale to the significantly increased demand.
The US is a different story. As a fraction of the total available spectrum, and even as it is being industrially widened from sub-split to high-split configurations, the fact remains that US capacity is a more difficult challenge. Commencing with shelter-at-home requirements, US traffic grew sharply, seemingly overnight. Comcast has publicly shared data on the increases in traffic scale since COVID started, along with transparency about the level of investment and technological attention that prepared us for “Black Swan” scenarios like a pandemic. This enabled more effective management of the additional traffic growth delivered over the Data Over Cable Service Interface Specification (DOCSIS) broadband network. As this paper will ultimately show, by adding an upstream PMA focus to the existing PMA suite, we were able to boost upstream capacity by 36%, from 86 Mbps to 117 Mbps.
Given these extraordinary circumstances, with the COVID crisis in full swing, and with the shift in internet usage, we refocused this paper to share our accelerated efforts in developing and deploying PMA for the US using DOCSIS 3.0 technology. Fortunately, the technology that was brought to bear on the challenges of US capacity was already under development. The effort was shaped by the early concepts found in CableLabs member publications from the late ‘90s into the early 2000s. Combined with state-of-the-art methods, including scaled cloud-based compute, and machine learning techniques such as Reinforcement Learning (RL), we were able to ensure system stability and optimize the network bandwidth as spectral conditions and demand changed.