ONF Connect 2018

ONOS is built from the ground up as a distributed-first SDN controller, with scalability and fault-tolerance as key attributes. Safely and efficiently managing software-defined networks in a distributed controller requires a wide array of complex distributed systems protocols that provide unique and often contrasting consistency and performance guarantees, making many off-the-shelf solutions inadequate for ONOS. Because of this, ONOS has long shipped with various custom implementations of distributed protocols and primitives developed as part of its core. But these tools are broadly applicable to distributed systems outside of software-defined networking, as well, so over the last two years an extensive effort has been underway to revamp and repackage them for general consumption by the open source community. The realization of that effort is Atomix 3.

Atomix 3 is a new stand-alone reactive framework for building fault tolerant distributed systems. The culmination of 5+ years of distributed systems development in both Atomix and ONOS core, it generalizes and abstracts the proven distributed systems primitives and protocols used in ONOS. The Owl release is the first ONOS release to ship with a new cluster architecture based on Atomix 3.
In this presentation, Jordan Halterman will provide an overview of Atomix 3 and the features used to achieve fault tolerance and scalability in ONOS. The talk will focus on the abstraction and generalization of ONOS distributed systems primitives and protocols in Atomix, present the new ONOS cluster architecture, and discuss how Atomix 3 facilitates service discovery, intra-cluster communication, replication, and coordination in ONOS 1.14.