Experiment with ArrayByteBufferPool performance

[gregw]

No overall size accounting
reserved buffer release always checks max memory
released buffers check max memory 1% of the time.
only a single thread can check memory at once.
single pass through buckets so no looping forever.

[gregw]

@lorban can you perf test this both in JMH and overall so we can see if this makes a difference, or do we need to look for a better approach.

[lorban]

I think this is the best we can do, given what we expect from ABBP.

Could you please just add the following test to the benchmark:

@Benchmark
    @BenchmarkMode({Mode.Throughput})
    public void fastPathAcquireRelease()
    {
        RetainableByteBuffer buffer = pool.acquire(65535, true);
        buffer.release();
    }

and modify the threads count to 32 in the main method?

[lorban]

FYI here's a perf run of this branch:
https://jenkins.webtide.net/job/load_testing/job/jetty-profiler-12.0.x/128/artifact/core/target/reports/CoreHandlerPerfTest_testNoGzipAsync/server/load-master-2/perf.hlog.html

Perf of the fast path is on par with 12.0.6.

[gregw]

@lorban what was the "hiccup" in the middle of that perf run?
Does it always do that?

[gregw]

@sbordet thoughts?

[lorban]

@gregw that hiccup is caused by async-profiler; disabling it makes the hiccup go away, and yes: it always dips around the 60th second.