Re: select/poll optimization

[Andrew Doran <ad%netbsd.org@localhost>]

On Thu, Feb 28, 2008 at 03:00:44PM +0000, Mindaugas Rasiukevicius wrote:

> As discussed with Andrew, here is the per-thread approach with the array of
> descriptors to store state:
> 
> http://www.netbsd.org/~rmind/perthread-select.diff

Based on the above patch, here's another that fixes some bugs and makes it
use the lightweight locking approach from the "per-CPU" patch. I have not
yet completed the changes to netsmb but otherwise it is complete.

        http://www.netbsd.org/~ad/combined-select.diff

With MySQL sysbench the above seems to buy about another 1.5% at peak on a 4
CPU box, and there is no lock contention internal to select() during the
test. Far from an ideal test but it does show that it works as intended.

        http://www.netbsd.org/~ad/select.png

Thanks,
Andrew