Difference between revisions of "HOWTO use ns-3 directly on the ORBIT testbed hardware"

Latest revision as of 21:57, 6 October 2014

Note: Some of the ns-3 documentation is stale below, as it references a program emu-udp-echo-client and emu-udp-echo-server that are not part of ns-3. The closest corresponding program in current ns-3 (as of ns-3.22) is src/fd-net-device/examples/fd-emu-udp-echo.cc.

We provide a realtime emulation package that allows us to connect ns-3 to real networks on real machines. Typically the real network will be a testbed of some kind. ORBIT is a two-tier laboratory emulator/field trial network project of WINLAB (Wireless Information Network Laboratory), at Rutgers. This wireless network emulator provides a large two-dimensional grid of 400 802.11 radio nodes as well as a number of smaller "sandbox" testbeds to allow one to test without reserving the main grid. This HOWTO shows how ns-3 scripts can be used to drive these radio nodes.

We assume that you have some experience with the ORBIT system. If you are new to ORBIT, please take a look at http://www.orbit-lab.org/ and go through the "Basic Tutorial" and the "Tutorials on controlling the testbed nodes" at a minimum. We will assume throughout this HOWTO that you have registered for an ORBIT account and have made a reservation on the ORBIT Scheduler for a testbed. This HOWTO assumes that you are on the sandbox one (sb1) testbed.

HOWTO use ns-3 directly on the ORBIT testbed hardware

We provide a node image on the ORBIT system that includes everything you need to get an ns-3 environment up and running on your testbed nodes. This includes the GNU toolchain, a copy of a precompiled fully-loaded ns-3.5 repository with NSC, Python bindings, emacs editor, etc., as well as a couple of specialized scripts to demonstrate a client-server (UDP Echo Client and Server) pair talking between two sandbox nodes.

The first step is to get this environment up on the nodes in your sandbox. In ORBIT terminology, we need to "image the nodes."

1. Connect to the sandbox console (we are assuming sandbox one, you need to substitute the correct address) and image the sandbox nodes:

   a.  ssh your-orbit-user-name@console.sb1.orbit-lab.org;
   b.  orbit load all ns-3.5.ndz 1200

This is a somewhat time-consuming process since you are actually copying disk images to the sandbox nodes. We have a relatively large image since it contains the entire GNU toolchain, ns-3, its reference traces, NSC, etc., and so we need to bump up the default timeout (with the 1200 second timeout specified in 1.a above). The example below took almost seventeen minutes to complete, so you must have patience!

You should see status messages appearing which indicate that the imaging process is proceeding as expected:

      Imaging nodes: 'system:topo:all' with image 'ns-3.5.ndz'
     (Domain:  default from hostname)
     (Timeout:  1200 sec.)
      INFO init: NodeHandler Version 4.4.0 (1921)
      INFO init: Experiment ID: sb1_2009_07_10_01_07_20
      INFO NodeHandler: Shutdown Flag Set - Switching all nodes Off...
      INFO Experiment: load system:exp:stdlib
      INFO prop.resetDelay: resetDelay = 210:Fixnum
      INFO prop.resetTries: resetTries = 1:Fixnum
      INFO Experiment: load system:exp:imageNode
      INFO prop.nodes: nodes = "system:topo:all":String
      INFO prop.image: image = "ns-3.5.ndz":String
      INFO prop.domain: domain = nil:NilClass
      INFO prop.timeout: timeout = 1200:Fixnum
      INFO Topology: Loading topology 'system:topo:all'.
      INFO stdlib: Waiting for nodes (Up/Down/Total): 0/2/2 - (still down: n_1_1,n_1_2)
      
      ...

      INFO whenAll: *: 'status[@value='UP']' fires
      INFO exp: Progress(0/0/2): 0/0/0 min(n_1_2)/avg/max (82) - Timeout: 1110 sec.
      INFO exp: Progress(0/0/2): 0/0/0 min(n_1_2)/avg/max (82) - Timeout: 1100 sec.
      INFO exp: Progress(0/0/2): 0/0/0 min(n_1_2)/avg/max (82) - Timeout: 1090 sec.
           
      ...
      
      INFO exp: Progress(1/0/2): 90/95/100 min(n_1_2)/avg/max (82) - Timeout: 225 sec.
      INFO exp: Progress(1/0/2): 90/95/100 min(n_1_2)/avg/max (82) - Timeout: 215 sec.
      INFO exp: Progress(2/0/2): 100/100/100 min()/avg/max (82) - Timeout: 205 sec.
      INFO exp:  -----------------------------
      INFO exp:  Imaging Process Done
      INFO exp:  - 2 node(s) succesfully imaged - See the topology file: 'system_topo_active_sb1.rb'
      INFO exp:  -----------------------------
      INFO Experiment: DONE!
      INFO ExecApp: Application 'commServer' finished
      INFO run: Experiment sb1_2009_07_13_02_47_05 finished after 16:46     

2. You can now take a look at the status of the nodes. Observe that they are in the POWEROFF state

   a.  orbit stat

       -----------------------------------------------
        INFO Topology: Loading topology 'system:topo:all'.
        Testbed : sb1
        Node n_1_1 - State: POWEROFF
        Node n_1_2 - State: POWEROFF
       -----------------------------------------------

3. You can now power up the nodes in the sandbox

   a.  orbit tell on all

       INFO Topology: Loading topology 'system:topo:all'.
       ---------------------------------------------------
        Testbed : sb1 - Command: on
        Node n_1_2 - Ok
        Node n_1_1 - Ok
       ---------------------------------------------------

4. Observe that the nodes are now powered up (it may take a while for the nodes to get around to get into the poweron state):

   e.  orbit stat

       -----------------------------------------------
        INFO Topology: Loading topology 'system:topo:all'.
        Testbed : sb1
        Node n_1_1 - State: POWERON
        Node n_1_2 - State: POWERON
       -----------------------------------------------

When you are done with this step, the sandbox nodes will have been powered up and will be in the process of booting. You now need to connect to the individual nodes in the sandbox and configure them. You may find it convenient to open a couple of terminal windows for this process. I will assume you do this, and in from now on, when I refer to [x,y] I mean the terminal window into which you you connect to the ORBIT node of array position [x,y].

You may see a "No route to host" error, when you try to ssh, until the nodes are actually up (done rebooting as a result of the power on step 3.a). You can ping -c 1 node1-1 and ping -c 1 node1-2 to verify they are up and responding before trying to ssh into them if you like.

Note that we configure "ath0" on node [1,1] to be an access point, and "ath0" on node [1,2] to be a station. We have to place the cards in promiscuous mode for the emu net device to work.

5. Connect to node [1,1] and configure its wireless. This will be the server.

   a.  ssh root@node1-1
   b.  wlanconfig ath0 destroy
   c.  wlanconfig ath0 create wlandev wifi0 wlanmode ap
   d.  ifconfig ath0 promisc up

3. Connect to node [1,2] and configure its wireless. This will be the client.

   a.  ssh root@node1-2
   b.  wlanconfig ath0 destroy
   c.  wlanconfig ath0 create wlandev wifi0 wlanmode sta
   d.  ifconfig ath0 promisc up

Now you can run the example ns-3 client server scripts on the testbed. We have configured [1,1] device "ath0" as an access point, and [1,2] device "ath0" as a station. Both devices must be in promiscuous mode. This can all be verified by using "ifconfig" and "iwconfig" on the appropriate console.

This is annoying, but for now we have got to manually set the hardware address of the interfaces in the ns-3 simulation script.

8. Find the hardware address of the wireless interface for the server script on node [1,1]

   a. ifconfig

Locate the output for the device ath0 in the output of ifconfig

   ath0      Link encap:Ethernet  HWaddr 06:60:b3:ac:2b:f5
             inet6 addr: fe80::460:b3ff:feac:2bf5/64 Scope:Link
             UP BROADCAST RUNNING PROMISC MULTICAST  MTU:1500  Metric:1
             RX packets:84 errors:0 dropped:0 overruns:0 frame:0
             TX packets:83 errors:0 dropped:0 overruns:0 carrier:0
             collisions:0 txqueuelen:0
             RX bytes:618 (618.0 B)  TX bytes:1710 (1.6 KiB)

Copy the HWaddr to the clipboard (in this case, 06:60:b3:ac:2b:f5)

9. Change the MAC address in the server script on node [1,1]

   a.  cd /home/tarballs/ns-allinone-3.5/ns-3.5/examples
   b.  emacs -nw emu-udp-echo-server.cc     # or use your favorite editor

Search for the MAC address assigmnet and replace the string with the HWaddr you just found, for example,

   ed->SetAddress (Mac48Address::ConvertFrom (Mac48Address ("00:00:00:00:00:01")));

becomes

   ed->SetAddress (Mac48Address::ConvertFrom (Mac48Address ("06:60:b3:ac:2b:f5")))'

10. Save the changes from your editor and build

   a.  cd ..
   b.  ./waf

   Waf: Entering directory `/home/tarballs/ns-allinone-3.5/ns-3.5/build'
   [609/710] cxx: examples/emu-udp-echo-server.cc -> build/debug/examples/emu-udp-echo-server_39.o
   [703/710] cxx_link: build/debug/examples/emu-udp-echo-server_39.o -> build/debug/examples/emu-udp-echo-server
   Waf: Leaving directory `/home/tarballs/ns-allinone-3.5/ns-3.5/build'
   'build' finished successfully (11.052s)

11. Find the hardware address of the wireless interface for the client script on node [1,2]

   a. ifconfig

Locate the output for the device ath0 in the output of ifconfig

   ath0      Link encap:Ethernet  HWaddr 06:60:b3:b0:c6:a4
             inet6 addr: fe80::460:b3ff:feb0:c6a4/64 Scope:Link
             UP BROADCAST RUNNING PROMISC MULTICAST  MTU:1500  Metric:1
             RX packets:0 errors:0 dropped:0 overruns:0 frame:0
             TX packets:1 errors:0 dropped:0 overruns:0 carrier:0
             collisions:0 txqueuelen:0
             RX bytes:0 (0.0 B)  TX bytes:70 (70.0 B)

Copy the HWaddr to the clipboard (in this case, 06:60:b3:b0:c6:a4)

12. Change the MAC address in the client script on node [1,2]

   a.  cd /home/tarballs/ns-allinone-3.5/ns-3.5/examples
   b.  emacs -nw emu-udp-echo-client.cc     # or use your favorite editor

Search for the MAC address assigmnet and replace the string with the HWaddr you just found, for example,

   ed->SetAddress (Mac48Address::ConvertFrom (Mac48Address ("00:00:00:00:00:02")));

becomes

   ed->SetAddress (Mac48Address::ConvertFrom (Mac48Address ("06:60:b3:b0:c6:a4")))'

13. Save the changes and build

   a.  cd ..
   b.  ./waf

   Waf: Entering directory `/home/tarballs/ns-allinone-3.5/ns-3.5/build'
   [608/710] cxx: examples/emu-udp-echo-client.cc -> build/debug/examples/emu-udp-echo-client_38.o
   [702/710] cxx_link: build/debug/examples/emu-udp-echo-client_38.o -> build/debug/examples/emu-udp-echo-client
   Waf: Leaving directory `/home/tarballs/ns-allinone-3.5/ns-3.5/build'
   'build' finished successfully (15.654s)

In this simple example, we are just going to manually start one ns-3 simulation script to act as a server for a trivial UDP echo application. Once the server is started on [1,1] for example, you can then go to the console for [1,2] and run the UPD echo client script. Let's do it. Remember you don't have all day to dawdle after running the server -- it exits automatically after 60 seconds.

14. Run the server on [1,1]

   a.  ./waf --run emu-udp-echo-server

15. Run the client on [1,2]

   a.  ./waf --run emu-udp-echo-client

Congratulations, you have now run your ns-3 simulation scripts on a testbed. That was exciting, wasn't it.

You can now look at the trace files on the client [1,2] and server [1,1] nodes and see what you've done. For example, to look at the pcap traces on the client node:

11. Display the pcap trace on [1,2]

   a. cd ..
   b. tcpdump -nn -tt -r emu-udp-echo-client-0-0.pcap  | more

Bonus HOWTO better integrate ns-3 with ORBIT

There are a number of ways that ns-3 could be further integrated with the ORBIT system. Patches are always welcome.

1. Use the ORBIT NodeHandler to drive your experiments: http://www.orbit-lab.org/wiki/Tutorial/HowToCommand

The ORBIT NodeHandler allows you to send shell commands to the various nodes in your experiment. You can use this facility to automate the steps you did manually in the HOWTO above.

2. Support the Orbit Measurement Library (OML): http://www.orbit-lab.org/wiki/Documentation/OML

Using OML, you need to write an XML description of the data you want to collect and compile it. You then get C code that you can link into your ns-3 script that will write your measurement data to the ORBIT database. This could be hooked into the ns-3 tracing framework. See http://www.orbit-lab.org/wiki/Tutorial/CollectingMeasurements

3. Athstats: http://www.orbit-lab.org/wiki/Documentation/Athstats

This is at its lowest level, a Linux program to query the device drivers on the nodes. You can use an ORBIT application to cause this information to be pulled out and stored in the ORBIT OML database during ns-3 driven experiments.

   frames transmitted successfully (at senders) 
   frames dropped due to excessive retries (at senders) 
   frames dropped due to FIFO errors (at senders) 
   frames dropped due to filtering (at senders) 
   short retries (for 802.11 RTS frames) (at senders) 
   long retries (for 802.11 DATA/MGMT frames) (at senders) 
   frames received successfully (at receivers) 
   frames dropped due to CRC errors (at receivers) 
   frames dropped due to FIFO errors (at receivers) 
   frames dropped due to PHY errors (at receivers) 

4. Use Aruba Sniffers http://www.orbit-lab.org/wiki/Documentation/ARuba

The ORBIT main grid has four Aruba sniffers at the four corners that can be configured to sniff specific channels and report every sniffed packet to the database. This gives the experimenters an independent framework to correlate 802.11 traffic during the course of the experiment for sanity checking as well as verifying the expected outcome of the experiment.

5. Write an ORBIT-specific net device

Right now we can control the settings of the node radios using the usual Linux tools. For example

 iwconfig ath0 txpower 30mW

There is a further level of integration with ORBIT available which would require another (ORBIT-specific) device to be written. The ORBIT folks have written a user-space library (libnet + libpcap + API) to allow you to write Ethernet packets, read full 802.11 frames, and get and set interface parameters. See http://www.orbit-lab.org/wiki/Documentation/Libmac

A new net device that uses libmac calls could integrate with the ns-3 Attribute System to allow control of the underlying ORBIT node hardware by the ns-3 script directly.

Craigdo 07:14, 13 July 2009 (UTC)