I had heard various tidbits here and there about Drizzle, including at MySQL2008, and had a pretty vague impression of how it was actually going to be different. The general theme seemed to be simpler, smaller, and faster.
If everything Brian described is true, it’s those things plus FUCKING AWESOME. I will join my friends (here, for example) in foaming at the mouth about it. Below is a rough list of things Brian mentioned that get me all hot’n'bothered about Drizzle, in no particular order:
- No Query Cache. The best part here is the reasoning for its absence: “If you’re relying on the query cache, you probably should have just used memcached to begin with.”
- In-Query Sharding Info. I’m not exactly sure of the details of the implementation, but it sounds like Drizzle will make it possible to spray queries through a to the right shard automatically.
- Pluggable Authentication. Finally. Authentication can also be completely turned off.
- Serializable Query Plans. This feature will allow the parser to be bypassed entirely on most queries. You simply send the query to MySQL, get the execution plan back, cache that, and send the execution plan back the next time you need the same query. That is Fucking Bad Ass (TM).
- Fewer Locks. Getting rid of a lot of the more advanced and rarely used features introduced in MySQL 5/5.1 (views, stored procedures, etc.), as well as some of the more basic stuff like authentication, has allowed Drizzle to lose 2/3 of the locks present in 5.0 (at least that’s what I think Brian said… sounds surreal…) which obviously opens the door for vast improvements on multicore architectures. It also sounded like Brian made the decision during his talk to discontinue MyISAM support in Drizzle to be able to get rid of another huge lock. I’m OK with it…
- Discontinued support for antique hardware. Lots of code ripped out because it’s no longer needed.
- Everything is pluggable, most things are optional. The Drizzle kernel is about 115kloc. Amazing.
Brian was very coy about benchmarks, leaving it to independent sources to run them, but it sounds like Drizzle will leave MySQL in the dust for most of the common applications seen in webapps. I can’t wait to try it and to use it on a few things.