By Amy Guo
Faculty members in the Stanford School of Engineering have launched the Platform Lab, an initiative that aims to develop new and more efficient ways of managing autonomous technology such as cars and drones for large-scale operations.
Within the lab, a team of Stanford faculty members and PhD students from across the school of engineering work together to create and develop new platforms, which are pieces of hardware or software that act as bases for constructing other types of technology.
Background and goals
The Platform Lab was established in 2015 when two smaller labs merged: the Stanford Experimental Data Center Laboratory and the Open Networking Research Center. The goal of the merger was to allow faculty in both labs to collaborate on larger research projects.
According to faculty director John Ousterhout, the members of the Platform Lab chose to focus their research on what they call “big control,” which is the ability to centralize management of very large swarms of devices. They hope to accomplish this by harnessing the power of data centers, which can manage thousands or even tens of thousands of machines working together on a particular application.
“Before the growth of the web, people tended to run applications on a single machine, [such as] your laptop or a server machine,” Ousterhout said. “But with the rise of sites like Google, Facebook and Youtube, which are supporting communities of millions of users, there’s no way you can run an application on just one machine and support all of those users.”
Currently, most devices are operated autonomously, performing different functions individually without a real central control to connect them. However, with data centers, the Platform Lab now faces the opportunity–and challenge–of using big control to run applications at an unprecedented scale.
To demonstrate the potential of big control, Ousterhout provided the example of commuting in the Bay Area 10 to 15 years from now, when there will be up to a million cars on the road, all of which will be self-driving. Rather than controlling traffic through each individual vehicle, he believes that a more efficient alternative is managing cars from a central location.
According to an article published by Stanford Engineering, a centralized model will make it possible for computers to keep track of millions of vehicles and plan ways around bottlenecks and hazards to efficiently and safely guide those vehicles to their various destinations.
However, a system that relies on a single central control center is not without its disadvantages. One of the biggest concerns is latency, or the time it takes for information to travel from a device to a data center and back again. According to Ousterhout, it would take about a tenth of a second for a car on the highway to communicate with a central data center and receive a response. However, in certain situations, such as when the car in front suddenly stops, the driver can’t afford to wait a tenth of a second for a command from a data center. As a result, Ousterhout concluded that certain decisions will have to be made purely locally, particularly those that have to be made with very low latency.
The potential of big control lies not only in traffic management. The Platform Lab is also looking to help control the “internet of things”– that is, those everyday devices such as refrigerators and thermostats which are connected to the internet. According to Stanford Engineering, centralized systems can also be used to coordinate large numbers of drones as they manage and move millions of packages within massive warehouses.
“The interesting thing about technology is it often develops in ways we can’t predict ahead of time,” Ousterhout said. “So there will probably be other kinds of devices that can benefit from this kind of system that we can’t even visualize today.”
The Platform Lab hopes that its research will eventually enable a very different world, using big control to make operations like disaster recovery and relief efforts more efficient and effective. In times of major disaster, big control could make it possible for relief teams to release thousands of drones to scour the area and report on the situation. Rather than individually specifying actions for each drone, humans will be able to command entire flights of drones simultaneously, leaving computer programs to determine the optimal course of action for each individual drone.
Ousterhout estimates that research projects typically take between five to 10 years to have practical impact, consisting of three to five years of research and additional time for commercialization.
While the Platform Lab is still relatively new and its research has not yet been commercialized, ideas from its predecessor labs have achieved wide success. One example exists in the field of Software Defined Networking, a current multi-billion dollar business that was originally developed and researched as part of the Open Networking and Research Center. Ousterhout expressed hope and optimism regarding the potential impact of the Platform Lab’s research.
“We’re all really excited about this, and we think it has the potential to be really impactful,” Ousterhout said. “One of the things I like about Stanford is that we’re exploring ideas that are so advanced that they’re crazy while building things that are practical and can have a real impact on the world. We come up with ideas that start off seeming nearly insane, but then end up becoming commonplace and used by millions of people around the world.”
Contact Amy Guo at acguo29 ‘at’ gmail.com.