For some time, networking capacity has been more of a bottleneck for online services than computer power. Today, Facebook Inc., which has some 2 billion users, is launching a new way to make sure sites and services can build the kind of networks needed to offer videos, photos and virtual reality experiences people are demanding.

Facebook’s answer is a new highly distributed network system, called Fabric Aggregator (pictured below), that’s composed of simple building blocks such as Facebook’s open-sourced Wedge 100 network switches, clustered in racks and connected in large numbers with several kinds of cable assemblies.

“At the scale more and more companies are at, you can’t just buy bigger and bigger switches to solve the problem,” Facebook engineering director Omar Baldonado said an interview ahead of the project’s introduction today at the Open Compute Project Summit in San Jose, California. “It’s like hitting the end of Moore’s Law on the networking side,” he added in a reference to the longstanding doubling of computer processor power every couple of years, which has recently slowed.

To contend with network demands that few other companies run into, Facebook has for several years been building what it calls a “disaggregated” network that breaks down traditional data centers into more smaller components. The benefits are that such systems can be scaled up more easily, the building blocks can be used in a variety of network roles and those smaller switches cause fewer problems if they fail.

In 2014, it developed a new switch called Wedge and software called the Facebook Open Switching System or FBOSS, which broke from the huge, monolithic designs of networking leaders such as Cisco Systems Inc. The next year, it released FBOSS into open source and released its proposed contribution of the Wedge to the Open Compute Project, which it helped form in 2011 as a way to standardize on new, more flexible designs for data center hardware. And in 2016, it introduced the Wedge 100 (below), its second-generation network switch, which was accepted into the OCP later that year. Now it’s donating the Fabric Aggregator layer to the OCP.

The crux of the Fabric Aggregator — fabric being a term for how large numbers of network switches are interconnected — is the cabling system, Baldonado said.

“Our solution is to break out of the chassis,” he said, essentially turning racks of switches in a data center into a functional chassis that can be connected by four distinct kinds of cabling configurations or “backplanes” (below).

“We want the whole industry to move to this disaggregated approach,” Baldonado said. That’s in no small part because Facebook would benefit from suppliers hewing to the standards it’s using in its own network.

Facebook is already using Fabric Aggregator in its own data centers.

Images: Facebook