On 5.1.2010, at 10.17, Stan Hoeppner wrote:
v1.1+ has somewhat faster search code. At least it's using boyer-moore with some of my own uglyness to make it support incremental searches. Wonder if there's a nicer and faster way to do that than what I implemented.
Hi Timo,
I'm not sure if you saw my original post. I'm using Dovecot 1.0.15 (Debian Lenny package) so I'm quite behind the current Dovecot rev.
I did, and I agree that it's Dovecot's searching code that's the bottleneck in your system, that's why I said it should work faster (with less CPU usage) in v1.1. You could get a v1.2 Debian package from backports.org.
And you're right, Dovecot doesn't try to use threads to do searching faster. I don't think it's all that useful in most installations, and implementing it is too much work.