For CSPs, unpredictability is the new normal in IP network evolution
CSPs have long had to battle with unforeseen network traffic events. The 2010 soccer World Cup, where the blare of the vuvuzelas prevented fans from making voice calls, causing them to overload stadium data networks. The 2016 Mirai botnet attack that took down Twitter, Netflix and Airbnb. The 2017 Chicago Pokémon GO festival, where cellular network congestion put the ‘slow’ in Slowpoke.
While radio access networks can struggle with sudden local traffic spikes, the critical IP networks that underpin the internet have always been built to handle the unexpected. But nobody was prepared for the scale of disruption brought by COVID-19, which swept away everyone’s traffic models, assumptions and predictions practically overnight.
Some CSPs saw an entire year’s predicted growth in just a few weeks, as housebound subscribers relied on their fixed and mobile connections for everything from videoconferencing, messaging and home-schooling to all-day Netflix and online gaming.
And with video calling putting an unanticipated strain on residential uplinks, CSPs had to take radical steps to keep the world connected – like removing data caps, blanket-upgrading home broadband connections, and even working with competitors to share traffic loads.
Will there ever be a “new normal” in IP networks?
While IP networks held up – the happy result of careful planning and a decade of investment – CSPs have been left wondering about the future. When will this be “over”? Will traffic ever return to “normal”? And if not, what kind of assumptions should drive network planning now?
The same questions are exercising industry analysts, too. The past eighteen months have seen such precipitous shifts that it’s hard to provide solid advice. “If you really put your finger on it, we’ve lost the ability to predict anything,” says Emma Mohr-McClune, service director at research and analytics firm Global Data. “We don’t know what the new normal will look like, or when it will come.”
In the absence of certainty, something has to fill the planning void. For Joe Madden, chief analyst at mobile market analyst firm Mobile Experts, it’s a case of focusing on what we do know – both about how networks and traffic have been evolving, and about human behavior.
It’s safe to assume that human usage will drive significant growth in mobile IP traffic, he says, and that it will continue to consist mostly of streaming video. As more people get faster connections, Mobile Experts is predicting a 100x increase in average data usage per user per month across fixed and mobile networks between 2021 and 2035.
On the Internet of Things (IoT) side, Madden says volumes will depend on the way enterprises handle video from connected devices. If video is mainly processed and analyzed at the network edge, IoT traffic over IP networks could be as low as 1.5GB per device per month. But if a portion of it is brought back to a central data center – to train machine learning models, for example – it could reach 67GB per device per month. Either way, he says, the rise of IoT data is not necessary to predict the 100x growth of overall mobile data.
What else can we say with confidence? One is that asymmetrical traffic models can be abandoned. Uplink-heavy video calling, videoconferencing and social live streaming are all on a growth trajectory, driving demand for symmetrical connections. The notion of the traditional “busy hour” is gone too, as home working means networks are busy with traffic all day.
These trends will spur convergence between residential fiber and 5G. “We need to plan for a more stable home working environment, and I’m sure we’ll see more solutions where 5G will be the automatic failover,” says Mohr-McClune. Madden takes it a step further. “Mobile broadband and fixed broadband will be terms that we throw away,” he says. “We’ll just call it broadband.”
Future IP traffic patterns remain unclear
So far, so predictable… sort of. We can assume that IP traffic will continue to grow, that video will continue to dominate, that fixed-mobile convergence will be increasingly critical, and that consumers will want fast and reliable upstream and downstream connectivity in the home.
But pitted against those broad assumptions is a vast array of uncertainties, making it impossible to predict with any degree of granularity how much traffic will cross IP networks, where it will come from, where it will be going, and what it will consist of.
For one thing, the pandemic has driven huge shifts in living and working patterns that may or may not become permanent. “There’s an element of unpredictability about where employees will be, because every enterprise will make its own decisions about home working policy,” says Mohr-McClune. “We know we’ll be using applications like Zoom and Teams, but where we’ll be using them is less clear.”
Complicating matters further is a possible flight from city living, as newly-remote workers pursue their dreams of moving closer to nature. How big an impact this will have, and how long-term a trend it will be, is currently anyone’s guess. But CSPs will need to make some guesses, because rural knowledge workers will need fast, reliable connectivity – which may mean revisiting rollout plans.
New applications and services will place unforeseen demands on the network
Then there’s the evolution of existing applications and services, and the emergence of new ones. The sudden shift to Zoom, Teams and other video tools was unexpected, but it’s reasonable to assume those tools will evolve to include more data-intensive features, like mixed-reality collaboration. The uncertainty over where users will be, however, makes it hard to predict the resulting impact on traffic.
Other trends are even harder to anticipate. In 2016, for example, the launch of Pokémon GO had an immediate and unforeseen impact on mobile networks. At the time, one European CSP reported that 7% of their users had used the game in a three-hour period, with the number of concurrent open sessions putting pressure on the network.
As 5G rolls out, similar consumer fads are likely to appear from nowhere, bringing sudden demand for capacity in places where it may not have been expected. “If you think about 5G social gaming, with groups of people playing the same AR/VR game in a single place, it’s really a mashup of applications,” says Mohr-McClune. “These mashups are going to be quite bandwidth-hungry”.
And while IoT traffic over transport networks may remain relatively small, there’s no guarantee that it won’t balloon in size. “A delivery drone may have to make decisions that aren’t easily decided by AI,” says Madden. “If it’s coming to a yard where kids are playing, it has to decide whether to land or not. That could require streaming video back to a central control center so a human can make real-time decisions.”
Security threats are hard to predict
And that’s just legitimate traffic. One of the big traffic trends of the early pandemic was a 40% rise in distributed denial of service (DDoS) attacks, often purchased as a service by gamers seeking to knock a rival offline. As the IoT vastly increases the potential attack surface, malicious traffic is likely to increase – but it’s unclear by how much, or what that traffic will look like.
What is certain is that individual DDoS attacks will get bigger, as attackers co-opt millions of unsecured IoT devices into botnets and DDoS-for-hire services proliferate online. In 2016, the Mirai botnet attack made history for breaching the 1 terabit-per-second barrier for the first time. Just four years later, Amazon reported an attack with a magnitude of 2.3Tbps.
With some attacks only lasting a few minutes, CSPs will need new strategies for identifying and mitigating these sudden and overwhelming floods of traffic. “We need to plan for more and harsher cyberattacks,” says Mohr-McClune. “That side of the market is only likely to become darker, more aggressive and better targeted – and that will drive new architecture choices.”
CSPs need a new definition of ‘worst-case scenario’
All of this means CSPs need to reconsider the stresses that might be put on IP networks in the future. “Operators have always planned for worst-case scenarios, but until the pandemic, we never really understood what the worst case could look like,” says Mohr-McClune. “We need to rebuild our assessment of what the worst case looks like, and plan towards that.”
In practice, that means ensuring the network is flexible enough to handle the revised view of what constitutes a worst-case scenario. While that will mean tough decisions around where and when to deploy radios and fiber, technological advances are making it progressively easier to ensure the IP network infrastructure can flex with demand.
“There are so many ways traffic is coming in now, and it’s going to so many endpoints,” says Vach Kompella, Vice President of IP Networks at Nokia. “We can’t design traffic models that are accurate, so we have to adjust as the models change. ”
Networks need more automation and architectural flexibility
Adjusting the network was once a job for human operators, but the complexity of modern networks makes manual handling of unforeseen events impossible.
“The more endpoints we have, the more services we have, the more connectivity that’s out there, the harder it is to have a central operations control site that’s monitoring the entire network with a bunch of humans,” says Kompella. “It gets very quickly into the realm of a lot of it having to be automated.”
Network automation is becoming much more advanced, he says, making IP networks self-aware and capable of automatically adjusting to handle unforeseen events like outages and DDoS attacks. Machine learning is starting to play a key role; for example by being able to automatically identify and reject DDoS traffic before it floods the network.
Architectural flexibility is crucial, too, requiring routers that can easily be reconfigured as new services emerge and demands evolve. “Programmability is very important,” says Kompella. “We don’t really know the direction that networks are going, so we need general-purpose or network-oriented processors, rather than fixed function ASICs.”
Network flexibility will allow CSPs to meet new customer demands
Flexibility in the IP network isn’t just about handling unexpected events and traffic patterns, but also about enabling CSPs to meet new customer demands. Kompella cites fixed-mobile convergence – where a person might reasonably expect to video-call from home, from the office, from the park or from a café – as one use case that will require a modern, automated IP network.
“When the end user doesn’t care what the access is, you have to design your network to offer the same set of services on any access,” he says. “And you have to be prepared that these services are going to be used from any access. It’s very important from an architecture point of view to realize this is coming, and that you have to optimize your deployment to be able to handle it.”
Mohr-McClune agrees, believing that workers – and their employers – will start to demand assurance from CSPs that they will be able to work effectively from anywhere. “We see operators focusing their future product development around an SLA that gives consumers the confidence to know that if they’re forced to work at home, then the home office will be an appropriately professional environment,” she says.
The new “normal” is anything but
So much has changed since the start of the pandemic that the idea of networks “returning” to any kind of pre-pandemic normal is long gone. With new working patterns, new applications, new services, new security threats and the imminent 5G-powered explosion of the industrial IoT, the only predictable thing now is that network traffic will become less and less predictable.
On the upside, we’ve come a long way from the 2010 World Cup, when CSPs had to manually reconfigure stadium networks to allow fans to communicate. As we move deeper into a “new abnormal”, automation, machine learning and architectural flexibility will allow networks to adapt automatically as traffic patterns shift in ways that nobody could have predicted.