This year, I only went to the sessions for OSCON. This is the sessions on Wednesday.
### [Big Bad PostgreSQL: A Case Study](http://conferences.oreillynet.com/cs/os2006/view/e_sess/8379), [Theo Schlossnagle](http://conferences.oreillynet.com/cs/os2006/view/e_spkr/1788)
This talk was about migrating some data warehouse systems from Oracle to PostgreSQL and the issues they ran into. Some of the issues were ones we had run into at work, about vacuuming being a hassle, no ability the upgrade between versions, no support for Oracle statistic functions like RANK OVER(), more limited partitioning support, and less evoled replication. The big advantage of PostgreSQL for them is that it could handle the complicated queries. And the Pl/PgSQL and Pl/Perl allowed them to do complicated stuff in the database.
### [Current State of the Linux Kernel](http://conferences.oreillynet.com/cs/os2006/view/e_sess/8387), [Greg Kroah-Hartman](http://conferences.oreillynet.com/cs/os2006/view/e_spkr/1644)
He talked about the Linux kernel. His talk had many points that I had read online. I found it impressive how much change the kernel undergoes and how well the development process holds together. The kernel is the poster child for distributed, evolving open source project producing something that works.
He also talked about the issues with external patches (like OpenAFS, Xen) and closed-source kernel modules. I think the new consensus that proprietary drivers are wrong and won’t be included in distros is a good thing. He pointed out that the common closed-source drivers, like VMware, NVIDIA, and ATI are operating in a gray area. They force people to compile the code. Anyone who distributes the resulting binaries is violating the GPL. Hopefully, this will force them to open source their drivers; I can’t see why VMware does not open-source their drivers. The graphics drivers will likely move into user space; there is no problem with proprietary X drivers.
### [The Atom Publishing Protocol as Universal Web Glue](http://conferences.oreillynet.com/cs/os2006/view/e_sess/9568), [Tim Bray](http://conferences.oreillynet.com/cs/os2006/view/e_spkr/814)
He described the new [Atom Publishing Protocol](http://bitworking.org/projects/atom/), showed some examples, and described a little bit how it is being used. The protocol is quite simple, basically a REST protocol for posting Atom entries and media files. Simple enough that I am now thinking about implementing a client for Palm, and a server for Typo.
I wish he had gone into more detail about the spec. Some history about other publishing protocols would have been nice. So would have some discusion about the design choices in the protocol.
### [Plagger: Pluggable RSS/Atom Aggregation](http://conferences.oreillynet.com/cs/os2006/view/e_sess/8993), [Tatsuhiko Miyagawa](http://conferences.oreillynet.com/cs/os2006/view/e_spkr/2734)
He described [Plagger](http://plagger.org/trac), a Perl framework for RSS/Atom aggregation. It has lots of plugins for pulling feeds, filtering them, and publishing to different formats.
I have been thinking that it would be interesting to do a personal feed aggregator web site. I have also been thinking about how to get my daily dose of news to my Palm.
### [Puppet: An Operating System Abstraction and Automation Framework](http://conferences.oreillynet.com/cs/os2006/view/e_spkr/2947), [Luke Kanies](http://conferences.oreillynet.com/cs/os2006/view/e_spkr/2947)
I was disappointed about this talk. I had investigated [Puppet](http://reductivelabs.com/projects/puppet/) at work and thought it was a great tool for doing system administration automation. He spent too much time talking about the purpose of the tool and the field with very little details. Technical details don’t work in a short talk but I got the impression that the audience wasn’t clear on what puppet is for and what it can do. Or why it is better than the competitors in this space like cfengine.
I think some concrete examples would have been better. Showing both the problem and the solution. A good example would have been setting up dozens of servers. First, the manual method. Second, the distributed scripts. Third, imaging. Finally, puppet can setup lots of machines in an automated way. Puppet can even take existing machines and change them to look like the declarated state.
### [Perl Lightning Talks](http://conferences.oreillynet.com/cs/os2006/view/e_sess/9222)
Luckily, I got to see Schwern’s play about module building. Very entertaining.
The rest of the talks were interesting with a few good ideas.