Yes, I'm still looking into this.
I agree that the zero-garbage code is difficult to follow, and Loom virtual
threads will have less reuse so our thread-locals will create more overhead
than they're worth in many cases. Fingers crossed for Valhalla to land before
Loom, but I haven't been follow
I see your point.
On Thu, Aug 26, 2021 at 11:21 AM Ralph Goers
wrote:
> While that is true it does mean that we are locking higher up in the call
> stack than we need to be.
>
> Ralph
>
> > On Aug 26, 2021, at 11:13 AM, Tim Perry wrote:
> >
> > I’m fairly certain the JIT will optimize out the l
While that is true it does mean that we are locking higher up in the call stack
than we need to be.
Ralph
> On Aug 26, 2021, at 11:13 AM, Tim Perry wrote:
>
> I’m fairly certain the JIT will optimize out the locking operations on the
> nested synchronized calls after a while. I’m not sure how
I’m fairly certain the JIT will optimize out the locking operations on the
nested synchronized calls after a while. I’m not sure how soon into the
performance tests that would happen.
Tim
> On Aug 26, 2021, at 10:55 AM, Ralph Goers wrote:
>
> So are you continuing to look at this? Frankly
So are you continuing to look at this? Frankly, the work on zero-gc stuff made
this
very complicated. I am not sure I can even follow the logic anymore. I also
noticed
that many of the methods dealing with byte buffers are synchronized. I am not
sure
why as only the method that moves data re
Sorry I missed replying to your question, Ralph!
> Did you add the no-op appender to Ceki’s project? Or are you using our
> NullAppender? I have
> my doubts about using that NullAppender as it does nothing - not even render
> the Layout, so
> it may get completely optimized away.
I added a no
I also tried that with similar results, which leads me to believe we're spending
too much time copying between buffers.
We've proven that log4j can get LogEvents to the appender very quickly and
efficiently.
Once we hit PatternLayout we write data to a StringBuilder.
AbstractStringBuilder used t
Ralph, maybe a dumb idea but... Can't we simply write to /dev/null to avoid
hardware's influence in the test results?
On Wed, Aug 25, 2021 at 8:05 PM Ralph Goers
wrote:
> Did you add the no-op appender to Ceki’s project? Or are you using our
> NullAppender? I have
> my doubts about using that Nu
