RFR: 8272807: Permit use of memory concurrent with pretouch
kim.barrett at oracle.com
Thu Aug 26 13:05:27 UTC 2021
> On Aug 26, 2021, at 2:59 AM, Aleksey Shipilev <shade at openjdk.java.net> wrote:
> On Thu, 26 Aug 2021 02:20:50 GMT, David Holmes <dholmes at openjdk.org> wrote:
>> That said, how can we be actively using an oop located in a page that we are also in the process of pre-touching?
> I think that is what this patch is supposed to enable: concurrent pretouch. Where "concurrent" might mean "concurrent with application code". For example, init the heap, let the Java code run, and then use a background GC worker to concurrently pre-touch the heap. ("Pre-" becomes a little fuzzy here…)
Among other uses, exactly.
And yes, “Pre-“ is a little fuzzy. I was originally intending to add a new “concurrent_touch_memory”
or something like that, but I’ve not been able to detect any performance difference between the old
and the new. I’ll be getting back to Aleksey on that later; I have more measurements I want to run.
More information about the hotspot-runtime-dev