EmergencyEMERGENCY? Get 24/7 Help Now!

Dagstuhl Seminar on Database Workload Management

 | August 7, 2012 |  Posted In: Tokutek, TokuView


A few weeks ago Bradley Kuszmaul and I attended the Dagstuhl Seminar on Database Workload Management.

The Dagstuhl computer science research center is (remotely) located in the countryside in Saarland, Germany. The actual building is an 18th Century Manor House, first retooled as an old-age home, and then a computer science research center. Workshop participants typically spend the whole week talking and working together.

Dagstuhl Computer Science Center

Shivnath Babu (Duke University), Goetz Graefe (Hewlett Packard), and Harumi Kuno (Hewlett Packard) did a great job organizing.  Slightly over half the participants came from industry and slightly over half from academia. (Bradley and I belong to both camps.)

I gave a talk on write-optimized data structures, such as Fractal Tree® indexes and LSM trees.  My point is that how one manages database workloads gets transformed with the availability of fast indexing. It was a great experience interacting with specialists and practitioners on platforms in addition to MySQL, and it brought home how many database problems are universal, transcending any particular platform.

Bradley spoke about Work Stealing and Cilk. I’ll let him discuss and post his talk.

On Thursday evening, the organizers continued the tradition of having a “gong show” session with 3-5 minute talks. In such a session, anything goes—purely humorous talks, open problems, futuristic or speculative talks. Somehow this relaxed format laden with humor means that people feel free to talk about ideas for which there is no other forum.

My gong-show presentation was tongue-in-cheek, but also technical.  I spoke about the scheduling aspects of workload management, and in particular, how you should schedule when you are a procrastinator. My talk wasn’t coming together until, Memento style, I decided to give the talk backwards.



  • I will guess that industry + academia means you did not invite anyone with experience running a DBMS in production. I think the academic side of the database field would do much better work if they learned more about production.

  • There were researchers and implementors. I don’t think there were any operations people per se, but I think the some of implementators are close enough to production to understand most of the reality that operations teams face.

    Also Mike and I didn’t organize the workshop, so we don’t get credit (or blame if you want to hand that out, but I would only hand out credit, since it was a good workshop).

Leave a Reply