DockerCon SF 2015: Networking Breakout
Download as PPTX, PDF6 likes6,232 views
This document provides an overview of Docker's new networking capabilities through libnetwork. It introduces libnetwork, which provides a pluggable driver-based networking stack for containers. Libnetwork implements the Container Network Model and provides APIs for creating and managing networks and endpoints. It supports multiple networking drivers like bridge and overlay. The goals are to make networking and services first-class objects in Docker and span networks across multiple hosts. The presentation encourages trying the new networking features in Docker experimental and contributing to libnetwork.
DockerCon SF 2015: Networking Breakout
- 1. Networking Breakout Madhu Venugopal Jana Radhakrishnan
- 2. Agenda • Introduction • Networking Deep Dive • Ecosystem • Q&A
- 3. Introduction
- 4. Why is Networking important? • Traditional Networking is incredibly vast and complex • Networking is an inherent part of distributed applications • Make it developer-friendly & application driven.
- 5. “We'll do for Networking, What Docker did for Compute.”
- 6. Goals • Make “network” & “service” as top-level objects • Provide a pluggable networking stack • Span networks across multiple hosts • Support multiple platforms
- 7. Libnetwork • Open Sourced in April • Over 200 Pull Requests • Over 200 GitHub Stars • Windows and FreeBSD ports in progress
- 8. Whats New? • Updated Networking Stack in Docker • Create Networks using the Docker CLI • Multi-host Networking • Services UI blue = experimental
- 10. What is Libnetwork • Library for creating and managing network stacks for containers • Test daemon/client called "dnet" • Driver-based networking • Implements the Container Network Model
- 11. Container Network Model (CNM) • Endpoint • Network • Sandbox
- 13. Libnetwork API • libnetwork.New • controller.ConfigureNetworkDriver • controller.NewNetwork • network.CreateEndpoint • endpoint.Join
- 14. RESTful API • Provides CRUD for Networks and Endpoints • /network • /network/<network_id>/endpoints • /network/<network_id>/endpoints/<endpoint_id> • /network/<network_id>/endpoints/<endpoint_id>/containers • /services • /services/<service_id> • /services/<service_id>/backends
- 15. Drivers • Drivers implement the Driver API • They provide the specifics of how a network and endpoint are implemented
- 16. Bridge Driver • Creates a Linux Bridge for each network • Creates a veth pair for each endpoint - One end is attached to the bridge - The other appears as eth0 inside the containers • iptables rules created for NAT
- 17. Overlay Driver • Creates a separate network namespace for every network - Facilitates overlapping IP address space across networks • Creates a Linux Bridge and VXLAN tunnels to every other discovered host • Creates a veth pair for each endpoint - One end is attached to the bridge - The other appears as eth0 inside the container • Network namespace connected to host network using NAT - Facilitates exiting the overlay network at every host(for external connectivity)
- 18. Network Plugins • Implemented using libnetwork's remote driver • Uses JSON-RPC transport • Can be written in any language • Can be deployed as a container
- 20. – R. Callon, RFC 1925 - The Twelve Networking Truths “One size never fits all.”
- 22. Call to Action! • Try the Docker Experimental Channel! - https://experimental.docker.com • Contribute to libnetwork - Raise an Issue or Submit a Pull Request • Chat with us on IRC - #docker-network on Freenode • Stop by at the booth for a demo
- 23. Q&A
- 24. Thank you Madhu Venugopal [email protected] @MadhuVenugopal Jana Radhakrishnan [email protected] @mrjana
Editor's Notes
- #18: Creates a separate network namespace for every network This is not the docker container namespace Facilitates overlapping IP address space across networks Creates a Linux Bridge and VXLAN tunnels to every other discovered host Linux bridge created inside the network namespace A single VNI is allocated globally for each network Creates a vxlan p2mp tunnel using that VNI and attaches that to the bridge Creates a veth pair for each endpoint One end is attached to the bridge The other appears as eth0 inside the container Network namespace connected to host network using NAT Facilitates exiting the overlay network at every host(for external connectivity)