wiki:GENIExperimenter/Tutorials/WiMAXOpenFlow/Design-Setup

Version 7 (modified by Ryan Izard, 8 years ago) (diff)

--

OpenFlow-Based Vertical Handoff over WiFi and WiMAX in the Orbit Testbed

Image Map

1. Design the Experiment

The goal of this tutorial is to perform a vertical handoff over WiFi and WiMAX in the Orbit testbed. Let's talk specifics with repect to each Orbit node. As briefly mentioned in the Overview, the experiment will be conducted with three Orbit nodes -- the client, the server, and the AP. Up first, we have the client node. The client is equipped with an Intel 6250 WiFi/WiMAX network adapter and an Atheros 5000 series WiFi network adapter. The Intel 6250 has a limitation – it can support the operation of either WiFi or WiMAX but not both simultaneously. As such, the Intel 6250 will serve as the WiMAX adapter and the Atheros 5000 series will serve as the WiFi adapter on the client. The OpenFlow-based handoff solution will choose which of these network interfaces it would like to use in order to relay messages to the server. Speaking of which, the server is equipped with the same wireless cards as the client. In order to receive packets from the client over both WiFi and WiMAX, the server will listen for the client's traffic over both its WiFi and WiMAX interfaces. To do this seamlessly, the server will also implement an OpenFlow-based solution. More details about how OpenFlow is used in both the server and the client are provided as the tutorial progresses. The third and final Orbit node necessary for this tutorial is the AP. Like the client and the server, the AP also contains both an Intel 6250 and an Atheros 5000 series wireless card; however, it will only use the Atheros card. The Atheros card on the AP is used by the software package hostapd to provide a WiFi network for both the client and the server machines. If you are interested in expanding upon this experiment, it is worth noting that Atheros cards play very well with hostapd-based networks; however, some Intel cards (such as the Intel 6250) are not directly supported. This is the reason we choose to use the Atheros over the Intel 6250 to provide the WiFi network.

Now, in order to provide a seamless handoff, the application should not be aware that the handoff takes place. The "seamlessness" of the handoff is still in development by the SDN team at Clemson. But for starters, from a networking point of view, if a network interface is brought up or is taken down, the IP address will be added to or removed from that network interface. This means the application's network socket will be broken if a break-before-make handoff occurs (which is oftentimes the case). When an application socket breaks, that means the application can no longer communicate over then network with it, and if this happens, the application must have some scheme to recreate the socket, otherwise the connection will be lost. In order to put as little reliance on the application layer as possible, a virtual tap interface is used. All applications will bind to the IP on this tap interface, and since tap interfaces are not physical, the only time it can “go down” is if we as programmers/network administrators take it down ourselves. Thus, it is a reliable way to make sure the application's socket stays “up”. In this tutorial, all application traffic will originate from and terminate at the tap interface.

The end-user will use the virtual tap interface (created by OpenVPN) for network connections, while the handoff execution will handle which physical interface to use. Now, the metric that decides which interface to use and when to perform the handoff is beyond the scope of this tutorial. Instead, this tutorial is designed to provide a basic framework for a handoff. To conduct the handoff, the Static Flow Pusher API in Floodlight, the OpenFlow controller, is used to insert OpenFlow flows manually, as determined by the handoff decision (e.g. a novel and super-cool metric a GENI experimenter has implemented). A Python script leverages the Static Flow Pusher API to add and remove flows.

The next logical question is where should this Static Flow Pusher insert the flows? In order to leverage the capabilities of OpenFlow, we need to have hardware and/or software that supports its use. Hence, Open vSwitch (OVS) is used in this tutorial. OVS is a software package that implements a virtual OpenFlow switch. This tutorial uses a pre-installed and pre-configured OVS network that incorporates the WiFi, WiMAX, and tap interfaces of the client as ports of virtual switches. The following is a picture depicting the general OVS topology used in this tutorial.

Each physical and tap interface has a corresponding OVS bridge. (In OVS terminology, an "OVS bridge" is essentially an "OpenFlow switch." It is called a bridge rather than a switch, since from the Linux kernel's perspective, the OVS bridge "bridges" the physical interface it is associated with.) These OVS bridges are linked together as if you actually plugged a physical switch into another with an Ethernet cable. The interfaces of the machine appear as ports on the OVS bridges, as do the internal links between the OVS bridges. With OVS, one can build a vitual network contained entirely within a single entity -- namely the client and server nodes of this tutorial. Each will employ internal OVS networks to alter the default behavior of the kernel with regard to networking. In essence, OVS allows the experimenter to circumvent the networking kernel and allow OpenFlow flows to route and modify packets instead.

2. Establish the Environment

  1. Using the Orbit images for the client, server, and AP, load each onto a separate node in an Orbit testbed. Make sure the node supports the hardware requirements for the experiment (Intel 6250 WiMAX and Atheros 5000 or 9000 series for the client and server, and Atheros 5000 or 9000 series for the AP). Boot these images after they are loaded and SSH into each machine.
  1. Remove the Forwarding module from the Floodlight OpenFlow controller. Floodlight uses what it calls a module loading system, where the user can write modules to perform a certain task or set of tasks. Each module can register for certain events. For example, the Forwarding module registers for PACKET_IN events where the controller is sent a packet from a connected switch. Upon such an event, the Forwarding module will send the packet out the correct port(s) depending on the destination. This module essentially implements a standard learning switch function where the OpenFlow-enabled switch behaves as if it were a standard network switch. We do not want this functionality, since we would like to have exclusive control over which port(s) our packets get forwarded.
    1. In the client and the server nodes, remove the Forwarding module.
    2. The module loading system maintains a list of the modules to be loaded at runtime. To remove the Forwarding module from this list (and thus disable it), open the /root/floodlight-0.90/src/main/resources/floodlightdefault.properties file and remove the line net.floodlightcontroller.forwarding.Forwarding,\.
    3. Browse to the root directory of the Floodlight project -- /root/floodlight-0.90. Execute ant in the terminal. ant is a Java-based build tool to build and compile Java projects. Upon success, ant will produce an updated executable jar file in the /root/floodlight-0.90/target directory named floodlight.jar.
  2. In the client, determine the MAC addresses of the WiMAX and WiFi network interface cards (NICs) for use in OpenFlow flows.
    1. Load the kernel modules for each of the NICs by executing modprobe i2400m_usb and modprobe ath5k, respectively.
    2. After the kernel modules are loaded, the devices should be visible in the system. Execute ifconfig wmx0 and ifconfig wlan1, respectively and note the MAC addresses of each NIC. These will be used in later steps.
      $ ifconfig wmx0
         wmx0      Link encap:Ethernet  HWaddr '''00:0c:29:04:5c:41'''  
             UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
             RX packets:0 errors:0 dropped:0 overruns:0 frame:0
             TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
             collisions:0 txqueuelen:1000 
             RX bytes:0 (0 KiB)  TX bytes:0 (0 KiB)
             Interrupt:18 Base address:0x1424 
      $ ifconfig wlan1
         wlan1      Link encap:Ethernet  HWaddr '''00:0c:29:04:5c:4b'''  
             UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
             RX packets:0 errors:0 dropped:0 overruns:0 frame:0
             TX packets:0 errors:0 dropped:0 overruns:0 carrier:0
             collisions:0 txqueuelen:1000 
             RX bytes:0 (0 KiB)  TX bytes:0 (0 KiB)
             Interrupt:19 Base address:0x14a4 
      
  3. Also in the client, customize the setup script gec18_setup.sh in /root/StartupScripts. This script is designed to (1) define user variables, (2) configure the tap interface with OpenVPN, (3) start Floodlight, (4) initialize and start OpenVswitch, and (5) configure Linux networking.
    1. Open the script with the text editor of your choice (vim, gedit, pico, nano, etc):
      $ gedit gec18_setup.sh
      
    2. There are numerous user defined variables at the top of the script. These are placeholders for commonly used system and configuration specific information throughout the script. We need to change a few of them to suit our needs for this tutorial. Modify the MAC address for IFACE_D_MAC to match that of the WiMAX interface noted from ifconfig wmx0. Also, modify the IFACE_bridge_E_IP variable to be an IP in the same subnet as the server node (yet to be configured). For the tutorial at GEC18, the server node will be configured for you and you will be assigned an IP to give your client node. Assign the IP you have been provided with to the IFACE_bridge_E_IP variable. Another variable that needs to be configured in the client is the MAC address, SSID, and channel of the AP node. The client needs these pieces of information in order to successfully associate with the AP. For this tutorial, the AP will also be pre-configured. Use the information about the AP that is provided.
      ###################
      #USR DEF VARIABLES#
      ###################
      
    3. Next, we need to remove any running instances of Floodlight and OVS. We do not want more than one instance of these processes running, otherwise chaos might ensue. So, a simple loop and kill works to terminate all running processes.
    4. Next, we need to create our tap interface. This is the network interface that will funnel all outbound packets from the applications on our client and send them into our OpenVswitch network.
      ###############
      #ADD TAP IFACE#
      ###############
      
      echo "OVPN: Installing tap interface, $IFACE_tap"
      openvpn --mktun --dev $IFACE_tap --lladdr 12:51:16:90:8f:ee
      
    5. After the tap interface, we need to start the Floodlight controller. The path to the controller executable generated by ant previosly is defined as the variable FL_path. Once Floodlight starts, it will not be connected to anything -- we haven't configured our OVS network yet. Once we do this, Floodlight will automatically detect and begin talking to the OVS bridges.
      ##################
      #START FLOODLIGHT#
      ##################
      
      echo "FL: Starting Floodlight..."
      cd / && ((java -jar ./root/floodlight/target/floodlight.jar) > floodlight-output 2>&1 &)
      echo "FL: Finished!"
      
    6. Next, we need to insert the Open vSwitch kernel module:
            ###################
            #START OPENVSWITCH#
            ###################
            
            echo "OVS: Configuring OVS..."
            echo "OVS: Checking for kernel module..."
            if [ -e $(lsmod | grep openvswitch) ]      
            then
      	   echo "OVS: ...inserting kernel module"
                 /sbin/rmmod bridge
                 sleep 1
      	   insmod /root/openvswitch-1.7.1/datapath/linux/openvswitch.ko
            else
      	   echo "OVS: ...kernel module already present"
            fi
      
    7. After that, we need to initialize Open vSwitch and its database:
      echo "OVS: Creating database"
      ovsdb-server --remote=punix:/usr/local/var/run/openvswitch/db.sock \
      --remote=db:Open_vSwitch,manager_options \
      --private-key=db:SSL,private_key \
      --certificate=db:SSL,certificate \
      --bootstrap-ca-cert=db:SSL,ca_cert \
      --pidfile --detach
      echo "OVS: Initializing OVS..."
      ovs-vsctl --no-wait init
      echo "OVS: Starting OVS..."
      ovs-vswitchd --pidfile --detach
      
    8. Now, we need to take down any pre-existing OVS bridges and create new ones. If we later try to create bridges with the same names, OVS will produce an error. There are options called --if-exists and --may-exists that are supposed to allow the creation of an OVS bridge/port under the condition it does not exist already. I have not been able to get these options to work, so we will use Bash scripting instead:
            echo "OVS: Removing any existing bridge, $IFACE_bridge_eth $IFACE_bridge_wlan0"
            echo "OVS: $IFACE_bridge_wimax $IFACE_bridge_int ..."
            
            ...
            
            ## Wlan0
            echo "OVS: Removing any existing bridge, $IFACE_bridge_wlan0..."
            if [ -n "$(ovs-vsctl show | grep $IFACE_bridge_wlan0)" ]      
            then
                 echo "OVS: ...removing $IFACE_bridge_wlan0"
                 ovs-vsctl del-br $IFACE_bridge_wlan0
            fi
            
            ## WiMAX
            if [ -n "$(ovs-vsctl show | grep $IFACE_bridge_wimax)" ]
            then
      	   echo "OVS: ...removing $IFACE_bridge_wimax"
      	   ovs-vsctl del-br $IFACE_bridge_wimax
            fi
            
            ## Internal
            if [ -n "$(ovs-vsctl show | grep $IFACE_bridge_int)" ]
            then
      	   echo "OVS: ...removing $IFACE_bridge_int"
      	   ovs-vsctl del-br $IFACE_bridge_int
            fi
      
            ######################
            #ADD INTERNAL BRIDGES#
            ######################
      
            ...
      
            ## Wlan0
            echo "OVS: Adding interface bridge, $IFACE_bridge_wlan0..."
            ovs-vsctl add-br $IFACE_bridge_wlan0
            echo "OVS: ...with port $IFACE_wlan0"
            ovs-vsctl add-port $IFACE_bridge_wlan0 $IFACE_wlan0 -- set Interface $IFACE_wlan0 ofport=$port_eth1
            echo "OVS: ...with port $IFACE_patch_wlan0_to_tap"
            ovs-vsctl add-port $IFACE_bridge_wlan0 $IFACE_patch_wlan0_to_tap -- set Interface $IFACE_patch_wlan0_to_tap ofport=$port_wlan0_to_tap
      
            ## WiMAX
            echo "OVS: Adding interface bridge, $IFACE_bridge_wimax..."
            ovs-vsctl add-br $IFACE_bridge_wimax
            echo "OVS: ...with port $IFACE_wimax"
            ovs-vsctl add-port $IFACE_bridge_wimax $IFACE_wimax -- set Interface $IFACE_wimax ofport=$port_eth2
            echo "OVS: ...with port $IFACE_patch_wimax_to_tap"
            ovs-vsctl add-port $IFACE_bridge_wimax $IFACE_patch_wimax_to_tap -- set Interface $IFACE_patch_wimax_to_tap ofport=$port_wimax_to_tap
      
            ## Internal
            echo "OVS: Adding interface bridge, $IFACE_bridge_int..."
            ovs-vsctl add-br $IFACE_bridge_int
            echo "OVS: ...with port $IFACE_tap"
            ovs-vsctl add-port $IFACE_bridge_int $IFACE_tap -- set Interface $IFACE_tap ofport=$port_tap0
            #echo "OVS: ...with port $IFACE_patch_tap_to_eth"
            #ovs-vsctl add-port $IFACE_bridge_int $IFACE_patch_tap_to_eth -- set Interface $IFACE_patch_tap_to_eth ofport=$port_tap_to_eth
            echo "OVS: ...with port $IFACE_patch_tap_to_wlan0"
            ovs-vsctl add-port $IFACE_bridge_int $IFACE_patch_tap_to_wlan0 -- set Interface $IFACE_patch_tap_to_wlan0 ofport=$port_tap_to_wlan0
            echo "OVS: ...with port $IFACE_patch_tap_to_wimax"
            ovs-vsctl add-port $IFACE_bridge_int $IFACE_patch_tap_to_wimax -- set Interface $IFACE_patch_tap_to_wimax ofport=$port_tap_to_wimax
      
    9. At this point, we're ready to set the patch ports between the OVS bridges. These create links between the OVS tap bridge and the OVS WiFi and WiMAX bridges in order to facilitate the flow of packets from the tap bridge to the physical interface of choice -- WiFi or WiMAX.
      #################
      #SET PATCH PORTS#
      #################
      
      ...
      
      echo "OVS: Patching ports $IFACE_patch_tap_to_wlan0, $IFACE_patch_wlan0_to_tap"
      ovs-vsctl set interface $IFACE_patch_tap_to_wlan0 type=patch
      ovs-vsctl set interface $IFACE_patch_tap_to_wlan0 options:peer=$IFACE_patch_wlan0_to_tap
      ovs-vsctl set interface $IFACE_patch_wlan0_to_tap type=patch
      ovs-vsctl set interface $IFACE_patch_wlan0_to_tap options:peer=$IFACE_patch_tap_to_wlan0
      
      echo "OVS: Patching ports $IFACE_patch_tap_to_wimax, $IFACE_patch_wimax_to_tap"
      ovs-vsctl set interface $IFACE_patch_tap_to_wimax type=patch
      ovs-vsctl set interface $IFACE_patch_tap_to_wimax options:peer=$IFACE_patch_wimax_to_tap
      ovs-vsctl set interface $IFACE_patch_wimax_to_tap type=patch
      ovs-vsctl set interface $IFACE_patch_wimax_to_tap options:peer=$IFACE_patch_tap_to_wimax
      
    10. Now, we need to assign each OVS bridge a unique ID (DPID) and point them to the address of the Floodlight controller. Floodlight will be run on the localhost, so the loopback address is defined within a the variable OVS_controllerIP and is set as the loopback address 127.0.0.1:6633. Port 6633 is the default OpenFlow port. As long as Floodlight and OVS are configured to use the same port number, any available port will suffice. We also need to tell the OVS bridges what to do if and when they lose connection to the Floodlight controller. The standalone mode allows them to default to a learning switch state if the link between the controller and the bridge is broken.
      ##########
      #SET DPID#
      ##########
      
      ...
      
      ## Set Wlan0 DPID
      echo "OVS: Setting $IFACE_bridge_wlan0 DPID to $OVS_switchDPID_wlan0..."
      ovs-vsctl set bridge $IFACE_bridge_wlan0 other-config:datapath-id=$OVS_switchDPID_wlan0
      
      ## Set WiMAX DPID
      echo "OVS: Setting $IFACE_bridge_wimax DPID to $OVS_switchDPID_wimax..."
      ovs-vsctl set bridge $IFACE_bridge_wimax other-config:datapath-id=$OVS_switchDPID_wimax
      
      ## Set Tap DPID
      echo "OVS: Setting $IFACE_bridge_int DPID to $OVS_switchDPID_tap..."
      ovs-vsctl set bridge $IFACE_bridge_int other-config:datapath-id=$OVS_switchDPID_tap
      
      ...
      
      ## Wlan0
      echo "OVS: Connecting $IFACE_bridge_wlan0 to controller at $OVS_controllerIP"
      ovs-vsctl set-controller $IFACE_bridge_wlan0 tcp:$OVS_controllerIP
      
      ## WiMAX
      echo "OVS: Connecting $IFACE_bridge_wimax to controller at $OVS_controllerIP"
      ovs-vsctl set-controller $IFACE_bridge_wimax tcp:$OVS_controllerIP
      
      ## Internal
      echo "OVS: Connecting $IFACE_bridge_int to controller at $OVS_controllerIP"
      ovs-vsctl set-controller $IFACE_bridge_int tcp:$OVS_controllerIP
      ovs-vsctl set bridge br_tap other-config:hwaddr=12:51:16:90:8f:ee
      
      echo "OVS: Finished!"
      
    11. Now that we have OVS and Floodlight running, the next thing to do in the setup script is to configure our network connections. We first need to disable IP forwarding, so that the kernel does not try to route packets around our OVS network. We then need to establish a link on each of the interface we would like to participate in the handoff -- WiFi and WiMAX. Note that WiFi is assigned to IFACE_B and WiMAX is assigned to IFACE_D. Also note the use of the WiFi AP variables we set in the first step of configuring this script. After WiFi and WiMAX each connect to their respective networks, we need to remove We also need to revoke the IPs from our physical interfaces (if any are present). What we want to do instead is have a single IP assigned to the tap OVS bridge on which applications can send and receive data. This will allow us to inject data/packets into our OVS network as well as pull packets destined for an application out of our OVS network.
      ##########################
      #CONFIGURE NETWORK ACCESS#
      ##########################
      
      ifconfig lo up
      
      ## Disable IP on physical interfaces
      echo "NTWK: Taking down $IFACE_wimax..."
      ifconfig $IFACE_wimax 0.0.0.0
      
      echo "NTWK: Taking down $IFACE_wlan0..."
      ifconfig $IFACE_wlan0 0.0.0.0
      
      ## Assign static IP addresses for bridge interfaces
      echo "NTWK: Assigning $IFACE_bridge_int IP as $IFACE_tap_IP..."
      ifconfig $IFACE_bridge_int $IFACE_tap_IP netmask 255.255.255.0
      
      ## Get an IP for bridge interfaces using DHCP
      echo "NTWK: Assigning $IFACE_bridge_wimax IP via dhclient..."
      dhclient $IFACE_bridge_wimax
      
      echo "NTWK: Assigning $IFACE_bridge_wlan0 IP via dhclient..."
      dhclient $IFACE_bridge_wlan0
      
      ...
      
      ## Turn off IP Forwarding
      echo "NTWK: Disabling IP Forwarding..."
      echo "0" > /proc/sys/net/ipv4/ip_forward
      
      ...     
      
      

  1. Save system_setup.sh and close your text editor.

Warnings

Warning Be on the lookout for typos in your scripts!

Notes

Note Write down your interface names, IP addresses, and subnet masks. All subnets must be the same for a Layer-2 handoff.

Tips

Tip If you need assistance, please ask for help!

Previous: Introduction

Next: Execute

Attachments (2)

Download all attachments as: .zip