Software Defined
Networking Tutorial
Shihabur R. Chowdhury
CS856 - Fall 2015
University of Waterloo
SDN Quick Recap
Traditional networks run distributed protocols to take
forwarding decisions
SDN has a centralized control plane that makes
forwarding decisions and asks the switches to act
according to that
VM Credentials
Username: sdn
Password: sdnpass
SDN and OpenFlow
Northbound API
(No standards yet)
Controller
(NOX, POX,
Floodlight etc.)
Southbound API
(OpenFlow, OnePK etc.)
OpenFlow Agent
packet
Datapath
SDN switch (pica8, indigo, Open vSwitch etc.)
OpenFlow
A switch specification and a switch to controller
communication protocol
Switches have forwarding tables
header (action,counter)
header:
source/destination IP
MAC
VLAN
TCP/UDP port etc.
header can have exact fields or wildcard fields
OpenFlow in Action
OpenFlow Controller
PacketIn
FlowMod
PacketOut
FeatureRequest
FeatureReply
FlowStat
StatReply
OpenFlow Switch
Ingress
Egress
Open vSwitch
An OpenFlow enabled virtual switch that can run on
commodity Linux machines
kernel module forwards the packet (data plane)
userspace module talks to the controller
A remote controller can control an OVS instance (control plane)
ovs-vsctl create/manage bridges
ovs-ofctl create/manage forwarding rules
But we need a network first !!
Mininet
De facto emulator for SDN
Uses Open vSwitch (ovs) to create SDN switches
Uses network namespaces to create hosts in their own
network namespace
Can emulate a whole network in one single machine
(even on a Raspberry pi)
Mininet Installation
Install mininet
sudo apt-get install mininet
Already installed in the VM
Show mininet options
mn -h
Start Mininet
Starting without any parameter creates a single switch
topology with two hosts connected with it and opens
mininet console
sudo mn
To view information about hosts and network use the
following commands
nodes, net, dump
Mininet Hosts
Hosts are processes running in their own network
namespace, i.e., hosts are processes with their own
network configuration
Run a command inside some host
h# command
h1 ifconfig
h1 ping -c 2 h2
Mininet
Open terminal to a host
Test network connectivity
pingall
Run an iperf between random pair of hosts
xterm h#
e.g., xterm h1
iperf
Set link bandwidth and delays
sudo mn --topo=single --link=tc,bw=10,delay=5ms
More Mininet
Python interpreter from Mininet terminal
Show the list of available methods in a host object
py dir(h1)
Show the IP address of a host
py
py h1.IP()
Set cpu usage limit for the hosts
sudo mn --topo=linear,3 --host=cfs,cpu=0.1
Mininet Built-in Topologies
Linear topology with 3 switches
Tree topology with depth 2
sudo mn --topo=linear,3 --switch ovsk
sudo mn --topo=tree,depth=2,fanout=2 --switch ovsk
Topology with a single switch
sudo mn --topo=single --switch ovsk
Working with OVS
Show details of switch s1
Show the flow rules in switch s1
ovs-ofctl dump-flows s1
Show port statistics in switch s1
ovs-ofctl show s1
ovs-ofctl dump-ports s1
Add a flow forwarding rule in switch s1
ovs-ofctl add-flow s1 <flow_spec>
Quick Exercise
Create a linear topology with 2 nodes
Open another terminal and dump flows in s1
Run iperf from mininet console
Dump the flows of s1 again
Dump the port statistics of s2
Mininet with Remote Controller
sudo mn --topo=single --controller=remote,
ip=127.0.0.1,port=6653
Try to ping h2 from h1
h1 ping h2
Manually Adding Flow Rules
There is currently no controller, therefore, no paths
Manually add a flow rule using ovs-ofctl
ovs-ofctl add-flow s1 in_port=1,action:output=2
ovs-ofctl add-flow s2 in_port=2,action:output=2
Mininet Python API
Mininet has a rich set of API in Python for creating your
own experiment
Create custom topologies, traffic patterns
Run applications inside hosts, etc.
Examples:
https://github.com/mininet/mininet/tree/master/examples
https://reproducingnetworkresearch.wordpress.com/
Mininet Python API Example
# Import mininet related packages
from mininet.net import Mininet
from mininet.node import Node, RemoteController
from mininet.log import setLogLevel, info
from mininet.node import CPULimitedHost
from mininet.link import TCLink
def run():
# Construct the network with cpu limited hosts and shaped links
net = Mininet(host = CPULimitedHost, link=TCLink)
# Create the network switches
s1, s2, s3 = [net.addSwitch(s) for s in 's1', 's2', 's3']
# Create the network hosts, each having 10% of the systems CPU
h1, h2, h3 = [net.addHost(h, cpu=0.1) for h in 'h1', 'h2', 'h3']
# Tell mininet to use a remote controller located at 127.0.0.1:6653
c1 = RemoteController('c1', ip='127.0.0.1', port=6653)
net.addController(c1)
# Add link between switches. Each link has a delay of 5ms and 10Mbps bandwidth
net.addLink(s1, s2, bw=10, delay='5ms')
net.addLink(s2, s3, bw=10, delay='5ms')
net.addLink(s3, s1, bw=10, delay='5ms')
Mininet Python API Example
# Add link between a host and a switch
for (h, s) in [(h1, s1), (h2, s2), (h3, s3)]:
net.addLink(h, s, bw=10, delay='10ms')
# Start each switch and assign it to the remote controller
for s in [s1, s2, s3]:
s.start([c1])
net.start()
# Start iperf server in h1
h1.cmd('iperf -s &')
# Run a iperf client on h2 and print the throughput
result = h2.cmd('iperf -yc -c ' + h1.IP() + ' -t 2').split(",")[-1]
print "Throughput between h1<-->h2: " + str(float(output)/1000000.0) + "Mbps"
net.stop()
if __name__ == '__main__':
setLogLevel('info')
run()
FlowVisor
A special OpenFlow controller that can slice the network
Allows multiple tenants to use the same physical network
FlowVisor Installation
Download flowvisor
Build
git clone git://github.
com/OPENNETWORKINGLAB/flowvisor.git
sudo apt-get install ant default-jdk build-essential
cd flowvisor && make
Install
sudo make install
FlowVisor Installation
Change directory ownership and permissions
sudo chown sdn:sdn -R /usr/local/share/db
sudo chmod -R 777 /usr/local/share/db
FlowVisor Configuration
Load the configuration file
Stop any running OpenFlow controller
Start flowvisor
sudo /etc/init.d/flowvisor start
Enable topology controller
sudo fvconfig load /etc/flowvisor/config.json
fvctl set-config --enable-topo-ctrl
Check configuration
fvctl get-config
Create Topology and Slices
s1
s2
s3
s4
h1
h2
h3
h4
Blue Slice
Red Slice
Create topology
Create a mininet topology
sudo mn --topo=linear,4 --arp --mac -controller=remote
Check the nodes and links from flowvisor
fvctl list-datapaths
fvctl list-links
Click network slices
Create two slices
fvctl add-slice blue tcp:127.0.0.1:7000 admin@blue
fvctl add-slice right tcp:127.0.0.1:8000 admin@red
List the slices
fvctl list-slices
Create flowspaces
Create flowspace partitions
fvctl
fvctl
fvctl
fvctl
fvctl
fvctl
add-flowspace
add-flowspace
add-flowspace
add-flowspace
add-flowspace
add-flowspace
dpid1 1 1 any blue=7
dpid2-p1 2 1 in_port=1
dpid2-p2 2 1 in_port=2
dpid4 4 1 any red=7
dpid3-p1 3 1 in_port=1
dpid3-p3 3 1 in_port=3
blue=7
blue=7
red=7
red=7
Run Controllers
Open two terminals
In terminal 1
sudo ovs-controller ptcp:7000
In terminal 2
sudo ovs-controller ptcp:8000
Test
In mininet console
h1 ping h2
h3 ping h4
h1 ping h3
