Data center traffic management is a big deal. Everything you’ve ever looked up on your phone has routed through a data center somewhere, through a massive version of a home router – the one you have to unplug, and plug back in every couple days.
Big data centers don’t have that luxury. When traffic gets backed up, a lot of bad things can happen. Data gets lost, requests time out, and work doesn’t get done. Because speed is even more important at a data center than it is for a home router, the algorithms that determine how traffic is handled are all hard-wired into the routers themselves.
But MIT researchers have proposed a new kind of router that could provide faster networks, and more adaptable traffic management for data centers.
“There are a lot of problems in computer networking we’ve never been able to solve at the speed that traffic actually flows through the network, because there wasn’t support directly in the network devices to analyze the traffic or act on the traffic as it arrives,” said Jennifer Rexford, a professor of computer science at Princeton speaking with MIT News.
These massive routers need to be fast, operating at speeds that put the fastest home Wi-Fi to shame, but they also need to be adaptable. These two aspects were previously irreconcilable, to be fast routers needed their control algorithms hard-wired, but to be adaptable they need to be programmable. Currently, when a new and improved control algorithm comes out, network engineers have to wait for new hardware to try it out in their data centers. They can’t just run a patch and update to the new version.
But researchers at MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) might have a way around those problems. Two papers unveiled this week detail methods which could allow network engineers to program data center routers without slowing down network traffic.
The first author listed on both papers, Anirudh Sivaraman, is an MIT grad student in electrical engineering, and his approach is a novel one. In the papers, he proposes a new router chip architecture which would allow for programmability, without sacrificing speed by keeping things simple.
Instead of building out a complex algorithm that can identify every type of network traffic and prioritize accordingly, Sivaraman’s approach relies on using simple circuit elements to perform basic operations in complex ways.
For instance, one circuit allows engineers to track running sums, while another performs packet scheduling, allowing engineers to change the way the circuit handles data on the fly – in response to real-time network conditions.
It might not seem like much, but by allowing even small changes to be made at the circuit level on data center routers, traffic management becomes much more adaptable. Which means better performance for data networks and faster speeds for the end user.
“What’s exciting about both of these works is that they really point to next-generation switch hardware that will be much, much more capable – and more importantly, more programmable, so that we can really change how the network functions without having to replace the equipment inside the network,” Rexford added.