RFR: Concurrent preclean

Aleksey Shipilev shade at redhat.com
Wed May 31 12:27:52 UTC 2017


On 05/31/2017 02:25 PM, Roman Kennke wrote:
> Just out of curiousity: would MT precleaning win us anything significant?

It probably would make the cycle shorter in wall time, which is better for
heuristics. But that's a second order effect, and probably does not worth
considering right now. The other thing is shorter gap to final mark, but that
again can be solved by draining SATB along the way. Let's see if that is
problematic in workloads.

-Aleksey



More information about the shenandoah-dev mailing list