The near future promises all sorts of exciting technologies, like autonomous vehicles, robot-powered remote surgery, and AI-driven manufacturing—yet these futuristic applications will rely on something stuck firmly in the past: network management.
Anyone who’s had a streaming movie buffer because a friend started downloading a video game, couldn’t call an Uber outside a concert because of an overloaded cellular network, or had a video chat freeze up mid-sentence might think twice before trusting these networks for their gallbladder surgery.
But, by the end of his five-year National Science Foundation-funded CAREER Award project, Feng Ye, an assistant professor of electrical and computer engineering at the University of Wisconsin-Madison, hopes to create network management tools that will make these and other network-dependent applications more responsive and more reliable.
Ye says that he believes there is more than enough network resources—mainly the capacity of a network to transfer data—available to meet current needs, as well as emerging AI-powered and advanced technologies. The problem is the way the resources are managed, which has not changed for a long time. This is especially true when it comes to edge networks, like those run by universities, large companies or institutions, in which data primarily flows between end users and the network via Wi-Fi routers or cell towers.
“It’s like building a road as wide as possible,” explains Ye, “while keeping the original traffic lights in place.”
Current network protocols are essentially reactive—they mete out bandwidth to devices based on a first-come, first-served basis, though they sometimes prioritize certain services if a signal is marked urgent. So if your smart thermostat wants to send a temperature reading to your phone and your TV wants to update its firmware in the middle of the day, that bandwidth use may make your important Zoom call a bit buggier than usual. That’s not too big of a deal—but if those same disruptions delayed a self-driving car’s decision-making by half a second, it could be deadly.
Today’s networks also have a bias toward optimizing download speeds versus uploading. While that currently makes sense in an environment where most users are consuming data and media rather than uploading lots of it (unless they are content creators or influencers), that download bias is a major bottleneck for emerging intelligent applications where computers and devices need to communicate back and forth with each other in near real-time.
To tweak this system, Ye plans to develop network management algorithms that look at network traffic proactively instead of reactively. That means AI-enabled management tools will simultaneously monitor data traffic, looking for recognizable patterns while also controlling the flow of data to eliminate delays as much as possible for all users, both those uploading and downloading data.
“The tool will forecast the traffic over a short but critical time window, e.g. the next second or 10 seconds, and decide which must be transmitted first—like driving or surgery instructions—and make sure the critical data reaches its destination before it expires,” says Ye.
Ye plans to produce these tools as open-source management tools. If they work well, he thinks they could be licensed for use in high-end network controllers, which would need powerful graphics processing units or AI processors to run the tools. Over time, however, he thinks it will be possible to produce routers running these network-management schemes inexpensively enough for everyday use in homes and businesses.
Ye also hopes by the end of his project to demonstrate his network-management technology using augmented and virtual reality. Virtual reality has yet to reach its full potential, he says, not because network bandwidth is too low, but primarily because of the limitations of current data management. By improving both upload and download speeds, he thinks his tools could finally make VR and AR viable at scale.
Widely deploying network management tools could also save money and resources. Next generation communications networks, Ye says, are probably more powerful and expensive than the average user will ever need. Proper network management could allow the current generation of networks to last for many more years, with newer, more expensive networks reserved for specialized users.
“Technically, we’re not trying to improve network capacity; I personally believe the infrastructure we have today is good enough for many use cases,” says Ye. “We’re just trying to better utilize the infrastructure that we already have.”
Top photo by Joel Hallberg