The next trip report is from Li-Wen Hsu:
I am very excited for having the chance to join the most important and largest annual BSD conference. Thanks to the FreeBSD Foundation, it's my first year to attend BSDCan. The main motivation for attending is that I'm in part of the project started by Craig Rodrigues, Jenkins CI for FreeBSD, and and I am honored to be invited join that group.
I arrived in Ottawa on May 13th. After checking into Residence and taking a short nap to ease jet-lag, I went to the Royal Oak Pub to join the pre-party of the developers. Sean Bruno quickly recognized me and introduced me to other developers. I talked with Steve Wills, Mark Linimon, Gavin Atkinson, and met Peter Wemm, my roommate.
The first day of the Developer Summit started with presentations about changes to the support plan and brainstorming about FreeBSD 11. During the break, I spoke to Mark Johnston, who completed the last piece of axge(4), our first USB 3.0 to gigabit ethernet adapter driver. It is written mainly by Kevin Lo and I provided some fixes in rx/tx routines. During our chat, we discussed the performance issues of axge(4) where he discovered there might be a limitation of calling rx/tx routines numbers per second in the USB stack. This is done by just a few lines of DTrace code. I was totally shocked by that and decided that I should learn more about it.
In the afternoon, I joined The Java working group where Greg Lewis introduced the history, current status, and we discussed the future plans of the Java port. We talked about how to improve user experience, support for important Java software, and the known problems of Java on FreeBSD. There was also a discussion on how to to get more developers who want to develop Java applications on FreeBSD. We think that DTrace support might be attractive for people who run Java on FreeBSD.
We had Thai food in the Hacker Lounge for dinner. George V. Neville-Neil and I talked about how to make more people develop or support their software on FreeBSD. I showed him Travis CI which is used by many open source projects developed on GitHub for their continuous integration needs. However, it cannot support FreeBSD in the near future. Being a ports committer for several years, I feel that many projects are willing to support FreeBSD, but lack the environment and experience. I think we should work more with external communities to address this. We discussed the possibility of establishing such a service for FreeBSD, following the successful model of RedPorts. There are many tricky parts and security issues to consider. Furthermore, the most important part is manpower. If any reader is interested in helping, please contact me.
On the second day of the Developer Summit, I went to the Continuous Integration and Testing with Jenkins for FreeBSD working group in the morning. In the first part of the meeting, Craig introduced Jenkins and how it is utilized in the FreeBSD.org cluster. He also covered the internal architecture of jenkins.FreeBSD.org. In the second part of the meeting, we discussed the next steps to work on this year. Craig and I helped Julio Merino set up a Jenkins instance on his laptop and Julio quickly hacked Kyua to let it generate JUnit output which can be parsed by Jenkins. This is very exciting to us because it means that we can have a trackable and easy-to-read continuous integration report. We believe this can help developers and contributors to produce higher quality code and to find items they can start to work on.
In the afternoon, I joined the Documentation Translation System Session where Benedict Reuschling introduced a process to translate documentation just like using gettext for software i18n and l10n. This process is done by translating docbook XML files to .po files with po4a, then translators can use their favorite po file editor to concentrate on the content instead of struggling with non-human readable XML files. It is also possible to establish a "translation memory" to remember the phrases and sentences that have already been translated for sharing between documents, which reduces duplicated work from translators. We also talked about a wish: one web-based system where casual translators can fix a translation by clicking a mouse while the backend takes care of the rest. The doc committers or another contributor can commit the change back to the doc repository.
Finally, I asked about continuous integration for the doc tree. Warren Block suggested that we can run igor for checking the errors, however there are some false positives that would bother people. During BSDCan, I joined two of the Doc Sprints. One night I asked Warren about "safe parameters" for igor and I quickly hacked igor to generate the checkstyle format XML output and pass it to the Jenkins checkstyle plugin. I presented the proof of concept on the second night. It is really great that people thought it is useful and encouraged me to setup it as a job on jenkins.FreeBSD.org. Warren will help me with this. In the future, this could also integrate with Phabricator as a "lint" tool for being as a filter.
I always wanted to revive the Traditional Chinese Document Translation Project. Fortunately, about two weeks after BSDCan, a volunteer sent a mail to the freebsd-doc mailing list stating that he wants to contribute to the Traditional Chinese translation of the FreeBSD Handbook. After discussing with him and with the help from people on EFNet/#bsddocs and doceng, I converted zh_TW from Big5 to UTF-8 in the doc tree for making future translation easier. This is really a good restart and I hope more people can join and we can have a complete Traditional Chinese handbook and other documents soon.
The following two days were the BSDCan sessions. The starting keynote speaker was Karl Lehenbauer from FlightAware. The most rememberable might be the slide "A billion dollars + Linus <= good people with a rigorous engineering process doing BSD." Also, there are some slides about FreeBSD tuning at FlightAware.
In Luigi Rizzo's talk In-kernel OpenvSwitch on FreeBSD, I learned a lot about how to port the Linux kernel subsystem to FreeBSD. In his work, he provides netlink sockets for the FreeBSD kernel. This is very good news because I always heard that people who are familiar with Linux want this feature.
Patrick Kelsey gave the libuinet talk. This is used to port the FreeBSD TCP/IP stack to userland. It means that the resource needed for a connection is in userspace memory and the kernel only needs to provide a packet interface, such as netmap. This is useful for an application that handles lots of concurrent connections. Patrick became a committer recently and I hope more work from him can bring libuinet closer to HEAD soon.
Pawel Jakub Dawidek and Mariusz Zaborski talked about their work on Capsicum and Casper. In this talk, they presented the lack of traditional security mechanism (setuid(2), chroot(2), P_SUGID, setrlimit(2)) provided by the system, and how an easier protection is provided by Capsicum. The Casper daemon provides services to sandboxed processess which do not have the necessary rights.
The FreeBSD Developer Summit and BSDCan overlap for one day. May 16 is the public track of the dev summit and I attended two sessions. Michal Dubiel from Semihalf gave the status update of OpenStack and OpenContrail on a FreeBSD host. I am glad that there are companies which invest in cloud technologies for FreeBSD. I hope this can be in production soon and maybe the FreeBSD cluster can have some setup to enrich developer resources. Another session I attended is lightweight reference counting, by Gleb Smirnoff. Using counter(9) in FreeBSD 10 as a reference count is really a brilliant idea. I am looking forward to seeing performance improve with this solution.
Arun Thomas gave a good tutorial of ARM and how BSD supports it. For a person like myself with no experience in embedded systems, it was a good start. Now I can have more fun with my Raspberry Pi.
Julio Merino talked about The FreeBSD Test Suite, which is really important to me and the FreeBSD continuous integration group. He also announced the plan to combine Kyua and Jenkins in the session. He hopes that the tests can be more complete and the CI pipeline can be more mature. There are still lots of things to be done!
Matt Ahrens presented the goals of the OpenZFS project and its current status. With this project, platforms like Illumos, FreeBSD, Linux, and OS X can directly interact with a shared, platform-independent ZFS code base. This will greatly reduce the effort to port changes between platforms and the tests can also be shared. The future of the OpenZFS features are amazing, including resumable zfs send/recv and device removal, which can make a system administrator's€™ life much easier.
The ASLR talk by Shawn Webb was awesome and this is definitely a feature that a paranoid person like myself hope will be merged to the main trunk soon. It seems to still have problems on ARM so he also asked for help from people with ARM experience.
On return home, I was surprised that Peter and I were on the same flight and sat next to each other. We talked about Linux containers and the projects that use it like Docker, which is the part that FreeBSD is not doing well. Currently, the resource limitation of the lightweight containers is not really complete. He said that the way we using servers, or the "computing nodes", are changing in lightspeed and we should not be left behind. We both agree that a modern operating system should put more effort in cloud and mobile solutions.
I would like to thank the FreeBSD Foundation again for sponsoring me to attend this great event. I made new friends and met people who only know each other on the Internet before. We shared many good ideas and it is really awesome to know there are so many people working on FreeBSD. I hope I can participate again next year!
