Open vSwitch - Part 2
A previous presentation in March 2013 at Bay Area Network Virtualization meetup covered the past, present, and predicted future of Open vSwitch. This talk picks up where that one left off, covering improvements made in Open vSwitch since then, new directions for the coming year, and some related work of interest in the industry.
About Ben Pfaff (twitter: @Ben_Pfaff)
Ben joined Nicira as one of its first employees in 2007 after finishing his PhD at Stanford. Since then he has been working on what became OpenFlow and Open vSwitch. He also made some early contributions to the NOX controller. He has been involved with free software since about 1996, when he started work on GNU PSPP and joined the Debian project.
More info @ http://meetup.com/openvswitch
Follow us on twitter @nvirters
2. What is Open vSwitch?
Semi-official description:
Open vSwitch is a production quality, multilayer virtual switch
licensed under the open source Apache 2.0 license. It is
designed to enable massive network automation through
programmatic extension, while still supporting standard
management interfaces and protocols (e.g. NetFlow, sFlow,
SPAN, RSPAN, CLI, LACP, 802.1ag).
The exciting parts:
● Write a program to control your network.
● Fast!
● Portable: OSes, hypervisors, CMSes, ...
3. Open vSwitch Hall
of Fame:
New Inductees
Alex Wang
Alexandru Copot
Alin Serdean
Andy Zhou
Ankur Sharma
Daniele Di Proietto
Flavio Leitner
Helmut Schaa
Jean Tourrilhes
Lorand Jakab
Nithin Raju
Pavithra Ramesh
Ryan Wilson
Thomas Graf
YAMAMOTO Takashi
Contributors with 10
or more commits
now (but not in
March 2013)
5. New Features
● Six major releases: v1.10 through v2.3.
● OpenFlow 1.1, 1.2, 1.3, some 1.4, prototypes for 1.5
● Tunnels: VXLAN, LISP, Geneve
● MPLS
● Multicast snooping
● RSTP
● TCP flags matching
9. ● 99% of traffic is not a problem:
– Small number of long flows
– Large number of medium-length flows.
● Tuning solves some problems.
● Real problem is large numbers of short flows:
– Port scans
– Peer-to-peer rendezvous servers
– Distributed systems
– Network monitoring applications
Performance, circa OVS 1.9
10. Megaflows (OVS v1.11)
● Most of the time the whole microflow doesn't
matter, e.g. MAC learning
● Push classifier into kernel
● Hard part: userspace generates megaflows
● Exact-match cache layer
12. DPDK/netmap/PF_RING/...
● What are they?
● How do they help?
– Fast and clever.
– No baggage.
– No ring transition.
● OVS v2.4 will support DPDK
– Basic work was simple
– Preliminary numbers are very good
13. Performance: Fairness
● Which flows get dropped?
– Random is bad.
– Per-tenant fairness is better.
● Per-port fairness (OVS v1.11).
● Per-destination fairness: no plans.
14. OVS 2014 Fall Conference
● Nov. 17 and 18 at VMware in Palo Alto
● Soliciting talks now until Oct. 6
– User, dev, admin, research talks all welcome
● Sign up to attend now
● More information:
– Follow link from openvswitch.org
● Free!