This is the second year I've attended Google's conference on scalability, the talks were more academic this time around. One organizer indicated that this was indeed the goal of this year's conference. I enjoyed last year how much of the information was really hands on, but there's some value in seeing what's coming down the research pipe.
The conference is an example of what googlers do with their "20% time".
I took lots of notes, but I'll summarize just the overarching concepts:
* Go Scale Yourself: Everyone seems to suffer from interruptions and it can take up to 1/2 hour to refocus after one. Asking the interrupter to file a ticket was suggested (something I've had lackluster success at). A better idea: keeping a pad of post it notes and writing down their request and telling them you'll get to it. The physical demonstration of accepting their request makes the requester feel they're being heard.
Good quote: "We're too busy mopping the floor to turn off the faucet." Sounds familiar.
* P2P: Lots of talk directly and indirectly on peer-to-peer, on the desktop, in the datacenter and under-water. Lots of interesting issues with lots of potential on both the small and large scale.
* Distributed Hashtables: Many talks mentioned distributed hashtables as a solution for one problem or another. I haven't read up on the details, so I don't have much to add.
* Multi-Core: We write a lot of plumbing code now to handle multiple threads, and really, we shouldn't have to. Cray's talk on Chapel showed how brittle our code is (because Doing It Right Is Hard) and Google had a presentation on Transactional Memory. If you've never heard of TM, then the talk is a great introduction.
I had an interesting chat with a database developer on the effects of solid state storage on database engines and a guy from Microsoft who had done a fair bit of P2P research.
The talks and slides will be up on YouTube over the next week. The talks were brought down a manageable 30 minutes from the originally scheduled hour, so if the presenters seem rushed, they probably were.
Monday, June 16, 2008
Google's Conference on Scalability
Subscribe to:
Post Comments (Atom)
0 comments:
Post a Comment