OpenConfig isn’t a standards body. It’s not even a formal group, really — more an invitation-only gathering via mailing list.
But with Google, Microsoft, AT&T, and BT Group among its instigators, OpenConfig might have a shot at making a difference with its project: an attempt at building an open, vendor-neutral model for network configuration and policy.
The key is that the OpenConfig members want to try this out now, showing the idea is implementable, rather than “be bogged down with the time it takes to standardize,” says Bikash Koley, Google’s principal network architect.
“The best way to standardize things is when you have pre-standard and proof-of-concept implementation, because then the debate can be: ‘Can this be better?’ versus ‘Can we do this thing?'” he says.
It’s all come together rapidly. Koley discussed the project in broad strokes in June. In recent weeks, OpenConfig has posted some models publicly on Github and one, a BGP configuration model, was submitted to the IETF in October. Koley thinks the group has already shown that this universal model isn’t intractable and that it really can aid in configuring networks. Formal standardization might come later, probably with in the IETF, he says.
Obviously, Koley sees some utility for Google in all this. He’s hoping OpenConfig’s network model could become a universal translator for configuring the network.
“The way we configure our SDN network is different from the way we configure our non-SDN network,” Koley says. The “SDN” way is a declarative, policy-based means of controlling the network — where an application tells the network what it wants rather than how to configure individual pieces of gear. (Policy-driven networking is also being developed by the Group-Based Policy efforts within OpenStack and the OpenDaylight Project.)
“We still can’t do that for the traditional vendor gear that we have quite a bit of in our network. Ultimately, we want to extend this to the vendor network,” he says.
The concept could be useful for other hyperscale data center operators — hence the participation of Microsoft. As networking becomes more policy-driven, OpenConfig’s models could also help telecom carriers by giving them a common way to communicate policy requirements across one another’s networks. (All the big carriers need to interconnect in order to reach users around the globe, or even in other U.S. states.)
The models are meant to be vendor-neutral, but that doesn’t mean vendors are getting cut out of the network. Just as vendors add their own tweaks to a standard such as BGP, they’ll be able to write extensions to whatever OpenConfig comes up with, Koley says.
OpenConfig is even considering extending to the optical layer, later, taking advantage of the standardized models already established by the International Telecommunications Union (ITU).
To YANG or Not To YANG
OpenConfig is using YANG as a starting point. Created for use with the Netconf protocol, YANG is a language for describing what’s in a network. In a sense, OpenConfig’s work is about outfitting YANG to suit the group’s purposes. “What it needed was a model for how you describe things in that language,” Koley says.
YANG has been around a long time and is often used for modeling a traditional network, one operated by an old command-line interface. There might be better options, as Mike Dvorkin — chief scientist of Insieme and now part of the Noiro Networks group within Cisco — pointed out on Twitter after our previous OpenConfig story ran. He noted he’s been using a combination of a modeling language and the Genie programming language.
Koley concedes YANG’s age but says it’s a good building block for its purposes. “You can also do very structure configuration definition, which is what we’re doing,” Koley says.