EA builds with changes to object monitor implementation to avoid pinning with virtual threads

[Andrii Lomakin]

Hi Alan.
Thank you for your comprehensive response.
I meant usage of io_uring in file read/write operations.
I can of course implement it myself using foreign function API but would be
cool it see once in JDK.

Thank you for your work on synchronized locks.
Once you wrote that support of synchronized split in two phases:
1. Simpler to implement but more expensive will be done first.
2. Then more complex but lightweight solution will provided.

Is it still the case ?

On Thu, 15 Feb 2024, 10:11 Alan Bateman, <Alan.Bateman at oracle.com> wrote:

> On 10/02/2024 12:24, Andrii Lomakin wrote:
> >
> > Hi Alan.
> > That is cool!
> >
> > Does this build support file async io aka io_uring features?
> >
>
> The purpose of these EA builds to have builds available with the changes
> to the object monitor implementation so we might get some feedback and
> help testing the changes.
>
> On file I/O, I'm not sure if you are asking about
> AsynchronousFileChannel implemented on io_uring or if you are asking
> about virtual threads doing file I/O using io_uring. There is a branch
> in the sandbox with prototype infrastructure but it will be some time
> before all the pieces can come together and properly evaluate for both
> of these cases. Right now, in these EA builds, doing file I/O in virtual
> thread when the file is opened with the options for synchronized I/O
> file/data integrity will increase the target parallelism for the
> duration of write operations. The latency of writes for these cases is
> significant. For direct I/O it is both read + write operations as these
> bypass the file caches.
>
> -Alan
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/loom-dev/attachments/20240217/9763fa82/attachment.htm>