This is a day-by-day account of the 1999 Ottawa Linux Symposium. Originally this was a set of entries in my diary for July 1999.
Thu Jul 22 1999: Spent the morning packing in preparation for Ottawa, and in the afternoon my parents drove me to Central Station to catch the train to Ottawa. After arriving at the Ottawa train station and catching a bus to the hotel, I mingled at the HP-sponsored reception for a bit over an hour, talked to an HP staffer about e-speak, peered at a disemboweled PA-RISC machine, and ate a vast number of delicious spring rolls.
Fri Jul 23 1999: Day 1. Four talks:
- The GNOME Mailer, Nat Friendman:
A reasonably well-done talk, though I don't recall anything new that I learned from it. - Simple DirectMedia Layer, Sam Lantinga:
Lantinga's company, Loki Productions, is working on Linux ports of several different games, and the DirectMedia Layer was written for this purpose. It provides a framebuffer abstraction that can sit on top of X11, or several of the full-screen graphics libraries available. It's interesting to see that, despite the recurring claims of X adding unacceptable overhead, it's quite possible to do fast graphics under X, and XFree86 4.0 is expected to improve things by better supporting accelerated cards. I don't plan on writing a game any time soon, and so this talk probably won't be of direct use to me, but it was still quite interesting. - The Bonobo Object Model, Miguel de Icaza:
Miguel seems to have mastered the art of the reality distortion field, borrowed from Steve Jobs; every time I see him talk about GNOME, I get all excited about it, forgetting how difficult GNOME is to compile out of CVS and how its development seems so unfocused. This talk was no exception; the OLE-inspired Bonobo object model seems very interesting, but it's impossible to tell how it works out in practice, since there's been so little implementation experience with it; implementation isn't even completed yet. - Freedom.net, Ian Goldberg:
My choice for the single best talk at OLS. Goldberg's company, Zero Knowledge, is working on a system for anonymous IP. The system has similarities to an anonymous remailer network, along with the onion routing and crowds research efforts. Take a central backbone of machines running the Freedom server software; users connect to one server, and negotiate a tunnel consisting of N servers, settling on a session key with each server in the tunnel. An outgoing IP packet is then encrypted N times, and sent to the first server, which removes one layer of encryption and passes it on to server #2. In the end, your packet emerges unencrypted from the final machine in the tunnel. Add some trickery to let packets find their way back, getting re-encrypted over the N return hops, and some application-level proxies, and you can now engage in Internet traffic without exposing your real IP address. Freedom.net would provide e-mail pseudonyms on top of this, with some safeguards to prevent spamming and for shutting down abusive nyms.Interesting though this all is, I'm not sure how ZK can possibly survive as a company. The plan seems to be to incite ISPs to run a Freedom server and make the anonymity service available to their customers. However, all experience indicates that the vast majority of people don't really care that much about online privacy; witness the stunning lack of reaction to the news that Microsoft Word embeds global user IDs in every document it creates. The people who do care about anonymity fall into two camps: cypherpunk-style activists who would use the system simply to show their support, and people with a real need for anonymity. The latter group subdivides further into legitimate users (whistleblowers, people under in totalitarian regimes) and a larger group of mischief-makers. Add them all up, and you don't have much of a market; I don't believe ISPs will be very interested. The Freedom network is a cool hack, and its operation makes for interesting discussion, but I can't imagine ZK being successful. (Yet they are funded, and plan to go public next year. Go figure.)
Rather than go to the Quake party, which sounded like it would be loud and obnoxious, I simply ate at a Thai restaurant near the hotel and spent the evening cranking through over 300 pages of my Ramsey Campbell novel.
Sat Jul 24 1999: Day 2. Five talks:
- Porting Linux to the ARM platforms, Russell M. King:
This was the only really disappointing talk I saw at OLS. Every so often I forget that hardware is really boring and read an article on a new chip architecture or attend a talk on porting Linux. Then I remember how dull memory caching and interrupt handling are, at least for a while. This talk spent a lot of time talking about the different lines of ARM chips and fiddly little bits of the porting effort, all fairly tedious topics that weren't helped by King's uncertain delivery. - Two Attempts at Server Optimization, Zach Brown:
The pivotal item here is that I learned a new way to structure a network server. There are two well-known ways: fork off a new process for each client, or use non-blocking I/O and handle all your clients in one process. The new way relies on Stephen Tweedie's SIGIO patches, which let you request that a signal be sent on every I/O event tied to a given file descriptor. Your signal handler gets a structure describing the nature of the event, and then has to implement a state machine so it can do the right thing depending on what the I/O event was (Stray thought: it would be interesting to experiment with this using the coroutine support in Christian Tismer's stackless Python.) This is close to my ideal talk: enough technical detail so that you learn something, but not so much that you get bogged down in trivia. - FreeS/WAN, Richard Briggs:
A high-level overview of the Linux IPSEC work being done in Ontario, with Henry Spencer as technical lead and Briggs as a primary implementor. If you didn't know IPSEC, this talk did very little to explain it to you and was probably rather vague and confusing, but someone who's read the IPSEC RFCs (like me) would have found this a useful status update. - Inside the Lizard, Mike Shaver and Mike Ang:
Covered the current status of Mozilla. A compulsivenetscape.*newsgroup watcher (again, like me) wouldn't have found much new here, because XUL and Necko and the XPCOM work are fairly old news. However, since most people seem to have no clue how Mozilla is doing, this talk probably changed some people's opinions (and one hopes it'll help kill off the incorrect conventional wisdom that Mozilla has somehow failed as a project). - Keynote, by Alan Cox:
AC talked about Brooks's law and how it only partly applies to free software projects. He attributed this to the ability to communicate directly between low-level implementors instead of filtering things through layers of management, and to the requirement to modularizing projects. (Free software projects are often modularized far more than is strictly necessary, and more than a corresponding commercial program would be.)
Overall OLS was really well thought out, and the quality level of talks was quite high; I saw only one dud, and two which were excellent. There were minor A/V glitches with microphones and projectors, but not fatal ones. I hope there will be a sequel to this conference next year.