Amid all the talk of software-defined networking (SDN), there are some management elements that have gotten overlooked, and Google‘s Bikash Koley wants to solicit industry help to create them — preferably in open-source, vendor-neutral forms.
To that end, Google will be opening up models it’s developed for network configuration and for describing network topology, the Google principal architect said Tuesday at Light Reading‘s Big Telecom Event in Chicago.
Google’s goal here is to create an abstracted view of the network, so that you’re programming the abstraction, rather than manipulating individual devices. That way, you can write software that programs an arbitrary network.
The result would be a declarative networking model, one where an operator tells the network what to do rather than how to do it.
Google wants something that’s more open, though — something that could apply to arbitrary vendors and networks. Vendors’ ideas of policy-based networking tend to center on creating something like an XML version of what’s already in their command-line interfaces, Koley told me after his talk. He was talking about vendors, plural, so it’s more than a dig at Cisco; the problem is the lack of a universal, vendor-neutral declarative model.
The endgame of all this would be to extend SDN beyond the network boundary. You can do this now with BGP, but you’re only sending “hints” rather than your full intent, Koley said.
By contrast, a declarative model would let you describe the network topology you want — where “topology” could even include instructions such as keeping certain traffic out of a certain region of the network.
“We believe that there’s an extremely high value that you can derive by essentially having more intent-driven network traffic exchange,” Koley said. “We would love to have ISP partners, as well as vendors, that are willing to work on this on an open way.”
So, Google has a couple of projects it wants to open up:
1. A network configuration model, one that’s vendor-neutral and extensible, and that can apply to “pretty much anything in the network,” Koley said. Yang provides the vocabulary for doing it, but there’s still no industrywide network model for it to describe. “It’s like saying I have a dictionary, but I don’t have any story,” he said.
Google has started building that “story” out of Yang and is “eager to work with the industry” on fleshing it out, Koley said.
2. A full Layer 0-7 topology model — that is, a consistent way of describing what’s in a network, from the fiber-optics up to the applications. As you might imagine, this is a harder problem.
One of the challenges, as Koley told me during a conference break, was to get the model to fit in a reasonable amount of space. (The problem becomes a little less difficult if you’re willing to transmit massive, director’s-cut-sized files every time you want to tell someone your topology.) The use of Protocol Buffers (a.k.a. protobuf), a data format Google previously created for servers, was key in creating a usable topology model, because it’s more capable than XML when it comes to transferring large chunks of data, Koley said.
An industry-standard topology model would make it possible to create useful network abstractions, which in turn would bring a programmable network closer to reality, Koley said.
Opening It All Up
Google wants to open both items as baselines for possible open-source management elements. The IETF is a likely venue for starting the dialogue and soliciting contributions, Koley said.
One thing Koley did not address — and I didn’t get a chance to ask about — are the group policy projects being developed in the OpenStack Foundation and the OpenDaylight Project, or the OpenStack Congress project to define policy for cloud services in general. These don’t precisely address the details Koley talked about — they certainly aren’t tackling that Layer 0-7 model, as far as I know — but if they remain open enough, they would be of help in creating the declarative network Koley talked about.