Usenix ATC

The Usenix Annual Technical Conference (ATC) used to be awesome, a word I don't use casually. My first was in Portland in 1985, and it had a series of talks, people, and great ideas I still ponder and use. We had two of these conferences every year, and it featured much of what was new in Unix, file systems, practical operating systems experiments, etc.

Over the years two things have happened to this conference. One, I think the field has gotten a little less varied and less rich in really interesting new ideas. Perhaps I am just getting jaded.

The other thing was Usenix's success with other conferences: security, file systems, LISA all covered with specificity parts of the original ATC track. Usenix has been eating its elder, and the ATC has been getting somewhat less interesting. I used to justify it and go to it easily twice a year. Now even the one conference doesn't always seem to be worth it.

The ATC this year is in San Diego, at the usual Town and Country. The size of the place impressed Andy Tannenbaum, but I am not a big fan. The sudden floral scents are nice, but the place is somewhat seedy under the vernier of nice drapes. It doesn't help that my ground floor room is directly across from the Pepsi machine.

The ATC had 200 submissions this year, a record. It also has 250 attendees, also a record I am told, though I suspect some of the first in Mel Ferentz's basement back in the Carter administration might have been smaller. We didn't come close to meeting our hotel block, with substantial penalties. It would have been cheaper for Usenix to cancel the conference. That contract and the current positions seem shortsighted on the part of the hotel, but I wouldn't be surprised if it were under new management in a little while.

Almost none of these observations are new to Usenix attendees and management. I have two marketing suggestions for Usenix:

I used to rate the Usenix conference this way: if I got one good idea out of it per day, it was worth it. (They even used that statement in some advertising once.) Usenix seems to be holding up on this. The paper quality seems to be quite good, something you might expect with a record number of submissions.

After a day and a half, the following have caught my notice:

The keynote, by James Hamilton of Amazon, was a lively presentation on optimizing the cost of large computer installations, focusing on power and air conditioning optimizations. He's an engineer with his eye on the ball. Best takaway: open the windows. (But put filters on: I thought of a very inconvenient attack if the outside air isn't filtered well enough.)

One of the best-paper winners was "Tolerating File-System Mistakes with EnvyFS" by Bairavasundaram et al from U Wisc (Madison). They divide a single file system stream to use three different file systems, and make sure that the results are the same. They aren't. Two benefits: debugging file systems (like ext3!), and they can ignore erroneous results by sticking with majority answers and make things more robust. I'd like to see it on FreeBSD. I don't want to use it, I want debugged file systems.