CheckMyTraffic

One of the most common mistakes that people make is to define a network topology in your experiment, only to have all the traffic traverse the shared control network instead. This is bad for you the experimenter because the traffic is neither shaped nor routed as the experiment topology specifies (the control net is a big 100Mb/1Gb LAN).

It is even worse for everyone else using the testbed because it can easily saturate the control network making services such as ssh, the web interface, or NFS slow or inaccessible. This leads to nastygrams from the operations staff telling you that your experiment is misconfigured and to please fix it.

Before you start any tests running, be sure to review this page and all its links to understand the problem.

After you start your tests, there are couple of things you can do to verify that traffic is doing what you think. From the "users" node you can run:

    portstats  

to get packet counters from the infrastructure switches. You should see traffic increasing only on the experimental interfaces.

On any individual node in the experiment, you can run "ps" to find the process ID of your application and then do:

    lsof -n -p 

and make sure that any network connections are using 10.x.x.x addresses rather than 155.98.x.x (the control network for Utah's Emulab) addresses.

Finally, on the node, you can use "ifconfig" and see the number of RX/TX bytes for each interface. Again, the counts should only be high on the 10.x.x.x interfaces.