This month, under the strain millions of people self-quarantined by COVID-19 have placed on broadband infrastructure, Netflix, Facebook, Disney, Microsoft, Sony, and YouTube agreed to temporarily reduce download speeds and video streaming quality in countries around the world. Nearly 90 out of the top 200 U.S. cities saw internet speeds decline in the past week, according to BroadbandNow. And Akamai found that global traffic on March 18 was running 67% higher than the typical daily average.
As a result of government and employer mandates to “shelter in place” and work remotely from home, internet subscribers are consuming more bandwidth than during the holidays and sporting events like the Super Bowl. At the same time, ISPs are under regulatory and consumer pressure to maintain a baseline quality of service. According to new research from Park Associates, 76% of households say it would be difficult to go without broadband. And in March, FCC chair Ajit Pai introduced the Keep Americans Connected Pledge, a telecom industry measure that asks companies to prioritize connectivity for essential services.
Internet service providers have taken steps to ensure that internet demand doesn’t overwhelm capacity. Beyond capital improvements, some — including Verizon, AT&T, Vodafone, Cox, and Telstra — are employing AI and machine learning to service networks strained by the traffic surges. Others aren’t — when reached for comment, Comcast, CenturyLink, and Fiber said they’re not using AI for network management.
Verizon told VentureBeat that it taps AI and machine learning to respond to shifts in usage, like the 75% increase in gaming traffic it saw from March 10 to 17.
“Analyzing patterns found in performance data, sensors, and alerting functions across all network platforms helps us identify performance issues before they impact the customer,” said a spokesperson via email. “For instance, based on analyzing patterns of performance, we are able to determine when parts of the network may need maintenance or replacement before a failure occurs and are able to roll a tech [person] to the scene or schedule that work into upcoming planned maintenance, saving on an extra trip.”
Verizon’s predictive algorithms monitor more than 4GB of data streaming every second from millions of network interfaces spanning everything from customers’ routers to sensors gathering temperature and weather data. The carrier’s analytics infrastructure allows it to predict 698 “customer-impacting” events before they happen and take steps to prevent them from occurring. On home networks, Verizon automates testing on a sample of over 60,000 in-home routers every two hours, ensuring that customers receive the speed of service they pay for — even with gaming, VPN, web traffic, and video traffic increasing 75%, 34%, 20%, and 12% week-over-week, respectively.
“We’re in an unprecedented situation,” said Verizon chief technology officer Kyle J. Malady. “We expect these peak hour percentages to fluctuate, so our engineers are continuing to closely monitor network usage patterns 24/7 and stand ready to adjust resources as changing demands arise. We continually evaluate peak data usage times and build our networks to stay ahead of that demand. While it is not clear yet how having millions of additional people working from home will impact usage patterns, we remain ready to address changes in demand, if needed.”
AT&T didn’t respond to VentureBeat for comment, but it previously said it uses a range of software-defined networking and network function virtualization technologies to mitigate spikes in network usage. One of these is what the company calls Enhanced Control, Orchestration, Management & Policy (ECOMP), which represents 8.5 million lines of code and supports over 100 different virtual network functions at all layers of AT&T’s network.
Like Verizon, AT&T applies predictive algorithms to its network to anticipate when hardware could potentially suffer downtime in the next days, weeks, or months. Historical analysis and pattern recognition also help optimize and route or reroute traffic. Separately, AT&T uses AI to manage its third-party cloud arrangements, such as with Microsoft and within its internal cloud and hybrid clouds. And on the mobile side, AI is aiding company technicians charged with spotting damage in cell towers from drone footage.
AT&T Labs vice president of advanced technology systems Mazin Gilbert says AT&T’s network-level AI can pick up signals indicating oncoming failures from vehicles in its repair fleet. In the future, he expects it’ll play a bigger role, potentially laying the groundwork for self-repairing systems.
“The network can’t be just software,” said Gilbert at the TM Forum Action Week conference in September 2019. “The network needs to be autonomous and pretty much zero-touch. It needs intelligence to know when it repairs itself, when it secures itself. The network needs to be contextual, personalized … [W]e have built these templates of intelligent agents. These are nothing more than closed-loop systems — closed-loop systems that capture data that can be configured for different problems. We push those in our network to collect data.”
Across the pond, Vodafone uses a cloud-based system called Neuron to generate network insights in real time. It’s built on top of Google Cloud with centralized access to data from over 600 servers in 11 countries, and it allows management to make decisions and take automated actions to improve service. For instance, Neuron can automatically assign more capacity in busy parts of the network while reducing capacity in parts that don’t require it.
Neuron is an evolution of a trial system Vodafone deployed to its mobile network in Germany with Huawei in 2017, dubbed Centralized Self-Organized Network (C-SON). C-SON identified the optimal settings to deliver voice over LTE services across 450 mobile cell sites chosen at random in four hours, a task that would have taken an engineer 2.5 months to perform manually. That same year, Vodafone’s Ireland subsidiary and Cisco teamed up to predict locations where 3G traffic will peak in the following hour, resulting in an average 6% improvement in mobile download speed and lower inference at the cell sites. And in Spain, Vodafone Spain piloted a system from Huawei and Ericsson that automatically chose the best frequency or node for each mobile connection.
“Neuron serves as the foundation for Vodafone’s data ocean and the brains of our business as we transform ourselves into a digital tech company,” said Vodafone group head of big data delivery Simon Harris. “Not only [can] we … gain real-time analytics capabilities across Vodafone products and services, [but we can] arrive at insights faster, which can then be used to offer more personalized product offerings to customers and to raise the bar on service.”
Vodafone — which reports that some of its networks have seen a 50% traffic uptick from the beginning of March — intends to use Neuron and other diagnostic tools to increase capacity where it’s needed and absorb new usage patterns. “Vodafone will be expanding capacity to manage this demand as much as possible,” said the company in a statement. “We also want to ensure that any congestion in the network does not negatively impact mission-critical and other essential communications during this period, such as for voice and digital access to health and education, or the ability for people to work from home.”
Cox, which serves 3.5 million internet subscribers in the U.S., says its management and service assurance strategy includes virtualizing portions of its network to “proactively and reactively” solve customer and network issues. The company’s software-defined networking capabilities tap AI to drive traffic optimization in the network backbone, delivering efficiencies in routing, latency, and resiliency in failover events.
“[W]e’re keeping a close eye at the individual node level to make sure we don’t approach any congestion thresholds and need to make any adjustments,” a spokesperson told VentureBeat. “Our focus is to help keep everyone connected during this unprecedented time, with remote workers and students learning from home top of mind … Similar to our normal process, if we see the network reach or exceed utilization thresholds, we will accelerate network upgrade plans in the impacted areas.”
Last year, Australian telecom provider Telstra began tapping AI to predict equipment failures on its network — and the company told IT News that it continues to do so. Telstra’s use of AI and machine learning extends to load balancing insofar as predictive models help reprioritize congested resources, ensuring customers on the network are minimally impacted.
Telstra, which said it would bring forward a $500 million capital expenditure from early 2021 to increase its network capacity during the pandemic, recently lifted data caps on home broadband customers until the end of April. “The data, which will be provided automatically, will help facilitate videoconferencing, voice over Wi-Fi, and cloud connectivity, all important tools when working from home or in isolation,” said CEO Andrew Penn in a statement.
By and large, ISPs that have deployed AI to manage COVID-19-related traffic surges are optimistic about the future. But they’re in uncharted waters.
According to broadband testing service Ooka, last week broadband speeds declined 4.9% from the previous week. Median download speeds dropped 38% in San Jose, California and 24% in New York. “Streaming platforms, telecom operators, and users, we all have a joint responsibility to take steps to ensure the smooth functioning of the internet during the battle against the virus propagation,” said Thierry Breton, the European Commission’s Internal Market Commissioner, in a statement.
AI might help — and already has helped — with respect to capacity. But it seems likely that for the foreseeable future, networks will be vulnerable to spikes in demand. AI can only do so much when what the world really needs is serious infrastructure investment.
View original article here Source