April 20, 2018

Guest Post by Ayaka Koshibe

At this year’s AsiaBSDCon, I presented a talk about a SDN network emulator called Mininet, and my ongoing work to make it more portable. That presentation was focused on the OpenBSD version of the port, and I breezed past the detail that I also had a version or Mininet working on FreeBSD. Because I was given the opportunity, I’d like to share a bit about the FreeBSD version of Mininet. It will not only be about what Mininet is and why it might be interesting, but also a recounting of my experience as a user making a first-time attempt at porting an application to FreeBSD.

First, a bit of context —

Thanks to its status as a buzzword, SDN, or software-defined networking, has different meanings to different people. The interpretation that I follow is that SDN is an approach to network design where its traffic-forwarding elements (datapaths or switches) expose a control channel through which a network control application with a global view of the network can steer it as if it were one entity. With this structure, managing the network means interacting with this logically central application, rather than dealing with each individual device and hoping that the network’s behavior converges to what you want.

Mininet started off as a tool used by academic researchers to emulate OpenFlow networks when they didn’t have convenient access to actual networks. Because of its history, Mininet became associated strongly with networks that use OpenFlow for their control channels. But, it has also become fairly popular among developers working in, and among several universities for research and teaching about, SDN.

I began using Mininet as an intern at my university’s network research lab. I was using FreeBSD by that time, and wasn’t too happy to learn that Mininet wouldn’t work on anything but Linux. I gradually got tired of having to run a Linux VM just to use Mininet, and one day it clicked in my mind that I can actually try porting it to FreeBSD.

From a user’s perspective, Mininet creates a SDN-capable network made of nodes and links. There are several types of nodes, including switches, the controller(s) that the switches are managed by, and end hosts. A developer can, for example, test their SDN application by replacing the controller with their own and running scripts on their hosts to send and receive various types of traffic, seeing how it copes.

Mininet creates a topology using the resource virtualization features that Linux has. Specifically, nodes are bash processes running in network namespaces, and the nodes are interconnected using veth virtual Ethernet links. Switches and controllers are just nodes whose shells have run the right commands to configure a software switch or start a controller application. Mininet can therefore be viewed as a series of Python libraries that run the system commands necessary to create network namespaces and veth interfaces, assemble a specified topology, and coordinate how user commands aimed at nodes (since they are just shells) are run.

Coming back to the port, I chose to use vnet jails to replace the network namespaces, and epair(4) links to replace the veth links. For the SDN functionality, I needed at least one switch and controller that can be run on FreeBSD. I chose OpenvSwitch(OVS) for the switch, since it was available in ports and is well-known by the SDN world, and the Ryu for the controller since it’s being actively developed and used and supports more recent versions of OpenFlow. (Mininet does give a user multiple choices of switches and controllers, but aside from OVS, none of them seemed to have been ported or well-tested outside of Linux). At the end of this exercise, I had what the creator of Mininet, Bob Lantz, would call “version 1”, which can be used to run customized networks and for development, but doesn’t support some of the fancier features like the drag-and-drop topology creator, or traffic-shaped links.

I have also been in touch with Bob, and have discussed the possibility of upstreaming my work. Although he was excited about it, I was asked about a script for creating VMs with Mininet preinstalled, and continuous integration support for my fork of the repository. I started taking a look at the release scripts for creating a VM, and after seeing that it would be much easier to use the scripts if I can get Mininet and Ryu added to the ports tree, I also tried a hand at submitting some ports. For CI support, Mininet uses Travis, which unfortunately doesn’t support FreeBSD. For this, I plan to look at a minimalistic CI tool called contbuild, which looks simple enough to get running and is written portably.

This is very much a work-in-progress, and one going at a glacial pace. Even though the company that I work for does use Mininet, but doesn’t use FreeBSD, so this is something that I’ve been working on in my free time. Earlier on, it was the learning curve that made progress slow. When I started, I hadn’t done anything more than run FreeBSD on a laptop, and uneventfully build a few applications from the ports tree. Right off the bat, using vnet jails meant learning how to build and run a custom kernel. This was the easy part, as the handbook was clear about how to do this. When I moved from using FreeBSD 10.3 to 11, I found that I can panic my machine by quickly creating and destroying OVS switches and jails. I submitted a bug report, but decided to go one step further and actually try to debug the panic for myself. With the help of a few people well-versed in systems programming and the developer’s handbook, I was able to come up with a fix, and get it accepted. This pretty much brings my porting experiment to the present day, where I’m slowly working out the pieces that I mentioned earlier.

In the beginning, I thought that this Mininet port would be a weekend project where I come out knowing thing or two about using vnet jails and with one less VM to run. Instead, it became a crash course in building and debugging kernels and submitting bug reports, patches, and ports. It’d like to mention that I wouldn’t have gotten far at all if it weren’t for the helpful folks, the documentation, and how debuggable FreeBSD is. I enjoy good challenges and learning experiences, and this has definitely been both.