Plumbers Conference retrospective
One of the reasons for this was that there was a strong presence from the kernel community, carried over from Kernel Summit. Since the purpose of Plumbers was to explore problems which span subsystems, having these folks in the room was a key factor. I’m told that it’s unlikely that the conferences will colocate next year, and I hope that it will succeed in drawing participation from kernel developers anyway.
There was a strong sense of cooperation among the different distributions, companies and projects which were represented, though less so between the kernel developers and userspace developers. These two groups would benefit from a better understanding of one another’s problems, and I hope that can be achieved through cross-participation in working events like Plumbers.
It’s common to picture the ecosystem as a stack or a sphere with some components at the bottom/center and others at the periphery, but these simplistic metaphors belie the complex and non-linear interdependencies which exist between projects. The kernel, the toolchain, the “plumbing”, applications, distributions, companies, and so on, don’t form a neat diagram, and each performs an essential function in making the overall ecosystem work.
I had a chance to talk briefly with Greg KH about his concerns and the way they were expressed, and have hope that some goodwill can be fostered there. I introduced him to Pete, who manages our kernel team, as a point of contact for a more nuanced dialog about our working relationship with the kernel community.
The discussions about the boot process were particularly interesting, as a great example of a problem which needs broad cooperation in order to solve effectively. For example, as a result of comparing the (quite different) bootcharts between Fedora and Ubuntu, developers from both distributions identified areas where significant gains were clearly possible without deep structural changes. Scott has isolated a long-standing issue which made our module loading sequence in Ubuntu much slower than it could be.
In between talks, I did some work on integrating apport with kerneloops. The result is that kernel oopses can be captured as Apport problem reports with full detail, and semi-automatically filed as bugs, in addition to being counted on kerneloops.org’s statistics. I’ve put an initial version into Ubuntu and sent the patch to Arjan for merging upstream, and we’re exploring the addition of kerneloops to our default installation to provide testing feedback to kernel developers from our users.