Intent-based Networking Systems (IBNS) may be the latest Gartner industry buzzword, but it captures something real that virtualization, SDN, and overlays have long sought: empowering network teams to operate more productively, at a higher level of abstraction.
In this talk, I’ll show how fast, accurate, and scalable network
modeling is not just a way to improve efficiency and uptime, but
helps transition to, and fully realize, intent-based
networking that is on the horizon.
First, I’ll motivate what network modeling is and what people use
it for, with use cases that benefit network operations teams,
including application connectivity troubleshooting, auditing,
change window verification, and others. I’ll then spend the
majority of time discussing practical challenges that arise when
building a comprehensive network modeling platform that enables
search, verification, and prediction at interactive speeds.
(1) The most basic challenge is establishing the intent
of existing devices, in a world of massive heterogeneity, across
vendors and device types. In some cases, the same
configuration from the same vendor yields different behavior across
different versions! How do you build accurate models without
building an army of network jockeys?
(2) A second challenge is making analysis work at scale,
especially in networks with 1000s of physical nodes, stateful
packet transformations, internet-scale tables, incomplete neighbor
discovery, and wacky firewall rules. The disk, memory, and
compute tradeoffs are often not what you’d expect, and
off-the-shelf processing and querying options are often not the
right fit for network data.
(3) The third challenge is one that may not have an obvious answer:
presenting data in understandable ways,
especially when modeling traffic aggregates, within and across data
centers, and covers layer 1 through 4, not just IP pairs at IP
level – and providing users with easy ways to describe their
At the end, I’ll describe work with customers to leverage network
modeling today. On one side of the spectrum, modeling is
helping identify errors before they manifest in a traditional
network that has grown organically over a decade. On the more
leading-edge side of the spectrum, a customer used APIs to help
build a fully verified intent-based network, one where the intent
is guaranteed to match the implementation, before the first packet
is even sent, and before any link fails the first time.
Co-Founder and CTO at Forward Networks
Brandon Heller received his PhD in Computer Science from Stanford University in June 2013. Involved in OpenFlow before it had that name, he served as main editor of the OpenFlow spec for three years. His academic projects included energy-efficient data centers (ElasticTree) and flexible network emulation (Mininet). As CTO at Forward Networks, he does whatever he can to help bring the world's best network modeling to to the world's network operators.