Saturday, September 8. 2007
my toys broke
Lum the Mad has a more complete write-up on Austin GDC — and the fact that it's not AGC — up on his blog. Naturally as a designer / producer he focused on different sessions than I did, but I agree with his overall assessment of the conference, and we did both go to Raph Koster's session.
Friday, September 7. 2007
Alive!
I survived Austin GDC (née AGC) pretty well. It's not nearly as hard as an attendee as for a volunteer (or someone staffing a booth). I very much enjoyed the technical sessions this year: I mostly attended talks on scaling the servers up to handle thousands of players, and gathering metric data.
Scaling continues to be a hot topic, with a surprising focus this year on databases. Latency seems like the big issue with them - and pretty much everyone seemed to have a custom cache layer between the database itself and the game servers, with some horror stories about games going live without them. The other big issue with databases is object serialization - how to store a character, all of the characters items, etc. in the database the "best" way. Nobody agrees on this one, naturally, because it's a fairly hard problem.
Interestingly, there was a professor (from Cornell?) and grad student who had very strong opinions on the subject of databases. Why can't game developers just read the literature? Why don't we realize object serialization is a solved problem? I'm interesting in doing more reading, but I suspect that latency is going to continue to be the bugbear here - what's considered acceptable performance in the wider database community is, I think, different.
Metrics (what are players doing? how often?) were also a big subject this year, up from the approximately one talk on the subject last year. There were a lot of interesting approaches discussed, and of course the big issue here is storage. One game, for instance (currently in beta) is generating 4.5 million database rows a week collecting data on game play.
Edit: The real tricky bit is storing objects in a database. Once you start using a database, suddenly people want to use the database like any other database: data mine it. The simple ways of serializing data (objects) into the database don't really let you also data mine, but setting it up for easy data mining tends to screw your performance.
Scaling continues to be a hot topic, with a surprising focus this year on databases. Latency seems like the big issue with them - and pretty much everyone seemed to have a custom cache layer between the database itself and the game servers, with some horror stories about games going live without them. The other big issue with databases is object serialization - how to store a character, all of the characters items, etc. in the database the "best" way. Nobody agrees on this one, naturally, because it's a fairly hard problem.
Interestingly, there was a professor (from Cornell?) and grad student who had very strong opinions on the subject of databases. Why can't game developers just read the literature? Why don't we realize object serialization is a solved problem? I'm interesting in doing more reading, but I suspect that latency is going to continue to be the bugbear here - what's considered acceptable performance in the wider database community is, I think, different.
Metrics (what are players doing? how often?) were also a big subject this year, up from the approximately one talk on the subject last year. There were a lot of interesting approaches discussed, and of course the big issue here is storage. One game, for instance (currently in beta) is generating 4.5 million database rows a week collecting data on game play.
Edit: The real tricky bit is storing objects in a database. Once you start using a database, suddenly people want to use the database like any other database: data mine it. The simple ways of serializing data (objects) into the database don't really let you also data mine, but setting it up for easy data mining tends to screw your performance.
(Page 1 of 1, totaling 2 entries)
